Pointers should not be `DATA`: they can never be stored in anything. (Function
arguments are not storage; they're like `let` bindings.)
Un-`SIZED` values may only live in the `Storage` storage class, so creating
pointers to them in other storage classes is meaningless.
The `ARGUMENT` flag should be set only on pointers in those storage classes that
are permitted to be passed to functions.
See comments in code for details.
Fixes#1513.
The WGSL front end knows that globals in the `Handle` storage class do not
produce references, but it attempts to manage this in two places which trip over
each other. As a consequence, referring to a `let` binding holding a pointer to
a global flips the variable's type from WGSL `ptr` to WGSL reference, with
bewildering consequences.
* [spv-in] Change shadow.spv test input to use StorageBuffer.
The ecosystem around Naga will generally not be able to use Vulkan adapters that
don't support the SPV_KHR_storage_buffer_storage_class (which was incorporated
into SPIR-V 1.3), so we can assume it is present.
Changing the test not to use runtime-sized arrays in the Uniform storage class
will allow Naga to tighten up some validation checks.
* [spv-in] Permit pointers to runtime arrays only in StorageBuffer.
Fixes#1519.
* [spv-in] Use `get_expr_handle` in `parse_image_write`
* [spv-in] Use BlockContext directly instead of its fields
* [spv-in] Use `get_expr_handle` for depth ref
The GLSL empty-global-name.frag test doesn't suffice because the GLSL front end
doesn't produce the same IR as the SPIR-V included in the bug report. As far as
I know, only a genuine SPIR-V input test can produce a global whose name is
`Some("")`.
Include the SPIR-V assembly source.
* Make default a switch case
Previously the default case of a switch statement was encoded as a block
in the statement but the wgsl spec defines it in such a way that the
default case ordering matters.
* [spv-out] Support for the new switch IR
* [dot-out] Use different labels for default cases
* Proof-of-concept for adding spans to validation errors.
Still missing: actually printing the damn stuff.
* Emit errors from analyzer in the CLI.
TODO: tests, I guess!
* Simplification refactoring: avoid avoiding allocations so vehemently.
* Mask helper traits with `as _`.
* Fix block iterator throwing up when span feature is disabled.
* Nest use statements.
* Add basic docs.
* Axe AddSpanResult.
Some compilers like shaderc introduce a full gl_PerVertex struct, this
includes gl_ClipDistance. Normally this isn't a problem since most
drivers optimize it away, but naga zero inits globals if they weren't
previously initialized. This causes gl_ClipDistance to be initialized to
zero which can be really bad for performance.
Minimize allocation in the namer. Heap-allocate string only when needed to
provide an owned key for a hash table, or to hold synthesized text. Try to reuse
allocations.
Many unnecessary trailing `_` characters are removed from test output. These
were all superfluous; separators are still inserted where necessary.
This is just a table mapping handles to their indices, but we can just fetch the
index directly from the handle. We never iterate over its entries, but even if
we did, we could at least use a HashSet instead of a HashMap.
