(and GL_NV_shader_subgroup_partitioned) based on GL/ES version
instead of predicating it on vulkan SPV generation
Also add AST testing.
The glsl.450.subgroup* files are largely the same as the spv.subgroup*
The glsl.es320.subgroup* files are the same as the 450 versions, but modified to be ES compatible.
Prior to this change, OES_primitive_bounding_box and EXT_primitive_bounding_box were both recognised as extensions, but only the name gl_BoundingBoxOES could be used. However the EXT version uses the name gl_BoundingBoxEXT instead. In addition, since GLES 3.2, the extension has been included in the core standard and the name gl_BoundingBox may be used instead. This change aims to make both extensions and the 3.2 core version all work.
One variable was only used in an 'assert' call. MSVC flagged this
as unused in Release. Suppress the warning and also add a static
cast to void so the variable becomes referenced.
The spvtools::Optimizer::Run method glslang is using constructs a default set
of spvtools::OptimizerOptions. This default set of options instructs the
validator to run. That is not quite correct since glslang will invoke the
validator _explicitly_ after the optimization pass.
Change-Id: I30f458304c6e7f81e89fc4ebd25eabbbd8348063
This is a better place for it logically, since it is not specific to
glsl->spirv translation. And in a future change I want to use it outside
of glslangtospv.
This is an alternate fix for the issue described in commit be63facd, whose
solution didn't work if there were non-trivial operations involved in computing
a constant initializer which caused the 'constant unfolding' code to kick in
(addConstantReferenceConversion). Instead, this change does the 'unfolding'
later in createSpvConstantFromConstUnionArray. If a reference-type constant has
survived that long, then folding is already done, this must be a 'real' (inside
a function) use of the constant, and it should be safe to unfold and apply the
bitcast.
Last year we changed 'volatile' to also act as 'coherent', but when I
resolved the memory model changes against that change I missed handling
volatile in a couple places that we check for coherent. There was also
a place in post-processing that acted as if the volatile memory access
flag has a literal number associated with it, when it doesn't.
Allow constructors to and from references to be constant folded. Section 4.3.3
says constructors whose arguments are all constant expressions must fold.
Disallow 'const' on buffer reference types. It is not a 'non-void transparent
basic data type' (it is not considered 'basic').
Handle buffer reference constants (which can be assigned to a non-const reference,
or can be further folded to another type of constant) by converting to
'constructor(uint64_t constant)' in addConversion.
Disallow == and != operators on reference types.
In decomposeIntrinsic a new TString was being allocated and passed into
a TVariable. That string was leaking. This CL converts the new TString
to call NewPoolTString to allocate from the TString pool.
Inside the grammar for function_identifier if the .function is null an
empty function name is allocated. This is allocated on the stack and
passed into TFunction as a pointer. TFunction just stores that pointer.
Later, when we access the name we will receive an invalid usage of a
stack allocated variable. This CL switches to using NewPoolTStringn for
the empty function name.
This fixes the comparison in macro body redefinitions, where initial
white-space differences do not matter, but internal white-space differences
do matter.
This has been a continually fragile area. Switching to a vector of real
objects, instead of a linearized stream of characters, removes a bunch of
code and generally makes this area more robust.
