platform/upstream/glslang.git
7 years agoReorder initializer fields to match variable declaration order (or vice versa) for...
Alex Szpakowski [Sun, 8 Jan 2017 22:21:17 +0000 (18:21 -0400)]
Reorder initializer fields to match variable declaration order (or vice versa) for several class constructors.

7 years agoAdd a virtual destructor to a class which has virtual methods.
Alex Szpakowski [Sun, 8 Jan 2017 21:57:21 +0000 (17:57 -0400)]
Add a virtual destructor to a class which has virtual methods.

7 years agoMark an overriden method in a subclass with ‘override’.
Alex Szpakowski [Sun, 8 Jan 2017 21:54:48 +0000 (17:54 -0400)]
Mark an overriden method in a subclass with ‘override’.

7 years agoNon-functional: White space after "//", mostly for copyrights.
John Kessenich [Fri, 6 Jan 2017 19:34:14 +0000 (12:34 -0700)]
Non-functional: White space after "//", mostly for copyrights.

7 years agoNon-Functional: Whitespace, comments, replace accidentally deleted comment.
John Kessenich [Fri, 6 Jan 2017 07:34:48 +0000 (00:34 -0700)]
Non-Functional: Whitespace, comments, replace accidentally deleted comment.

- fixed ParseHelper.cpp newlines (crlf -> lf)
- removed trailing white space in most source files
- fix some spelling issues
- extra blank lines
- tabs to spaces
- replace #include comment about no location

7 years agoMerge pull request #661 from KhronosGroup/fix-include-error-paths
John Kessenich [Fri, 6 Jan 2017 07:32:39 +0000 (00:32 -0700)]
Merge pull request #661 from KhronosGroup/fix-include-error-paths

PP: #include: simplify the different paths out of #include.

7 years agoPP: #include: simplify the different paths out of #include.
John Kessenich [Thu, 5 Jan 2017 20:32:52 +0000 (13:32 -0700)]
PP: #include: simplify the different paths out of #include.

- some paths didn't release 'res'
- token is always '\n' after proper acceptance of the directive itself,
  so no need to test it, change it to '\n', etc.
- assuming setCurrentColumn(0) is not needed unless there are header tokens,
  but not clear why it is ever needed

Note: much of the simplified code read as if the included header tokens had
actually been processed, versus queued up for processing; maybe that explains
some things.

7 years agoPP: #include: add names to error messages, so that lexical analysis can be tested.
John Kessenich [Thu, 5 Jan 2017 19:18:34 +0000 (12:18 -0700)]
PP: #include: add names to error messages, so that lexical analysis can be tested.

7 years agoNon-functional: Very minor clean up.
John Kessenich [Thu, 5 Jan 2017 17:45:32 +0000 (10:45 -0700)]
Non-functional: Very minor clean up.

7 years agoMerge pull request #650 from steve-lunarg/lvalue-swizzle-fix
John Kessenich [Thu, 5 Jan 2017 17:40:14 +0000 (10:40 -0700)]
Merge pull request #650 from steve-lunarg/lvalue-swizzle-fix

HLSL: allow destination swizzles when writing RWTexture/RWBuffer

7 years agoHLSL: Fix issue #658: Don't adopt initializer constness from declaration.
John Kessenich [Thu, 5 Jan 2017 17:28:26 +0000 (10:28 -0700)]
HLSL: Fix issue #658: Don't adopt initializer constness from declaration.

This also makes it match how GLSL handles the same thing.

7 years agoMerge pull request #648 from steve-lunarg/type-identifiers
John Kessenich [Wed, 4 Jan 2017 21:07:34 +0000 (14:07 -0700)]
Merge pull request #648 from steve-lunarg/type-identifiers

HLSL: allow type keywords as identifiers, and add half type

7 years agoMerge pull request #632 from steve-lunarg/structure-splitting
John Kessenich [Wed, 4 Jan 2017 18:41:36 +0000 (11:41 -0700)]
Merge pull request #632 from steve-lunarg/structure-splitting

HLSL: inter-stage structure splitting.

7 years agoMerge pull request #659 from steve-lunarg/d3dcolortoubyte4
John Kessenich [Tue, 3 Jan 2017 22:34:33 +0000 (15:34 -0700)]
Merge pull request #659 from steve-lunarg/d3dcolortoubyte4

Add D3DCOLORtoUBYTE4 decomposition

7 years agoMerge pull request #647 from steve-lunarg/default-fn-params
John Kessenich [Tue, 3 Jan 2017 22:30:05 +0000 (15:30 -0700)]
Merge pull request #647 from steve-lunarg/default-fn-params

HLSL: default function parameters

7 years agoAdd EOpD3DCOLORtoUBYTE4 decomposition
steve-lunarg [Tue, 3 Jan 2017 21:42:18 +0000 (14:42 -0700)]
Add EOpD3DCOLORtoUBYTE4 decomposition

7 years agoPP: Recognize <> style #include header names. I.e., #include <header-name>.
John Kessenich [Tue, 3 Jan 2017 03:12:08 +0000 (20:12 -0700)]
PP: Recognize <> style #include header names. I.e., #include <header-name>.

Also correctly test and handle missing newline.

7 years agoHLSL: allow destination swizzles when writing RWTexture/RWBuffer objects.
steve-lunarg [Wed, 28 Dec 2016 17:03:58 +0000 (10:03 -0700)]
HLSL: allow destination swizzles when writing RWTexture/RWBuffer objects.

Reads and write syntax to UAV objects is turned into EOpImageLoad/Store
operations.  This translation did not support destination swizzles,
for example, "mybuffer[tc].zyx = 3;", so such statements would fail to
compile.  Now they work.

Parial updates are explicitly prohibited.

New test: hlsl.rw.swizzle.frag

7 years agoPP: Fix issue #426, recover from bad-source macro expansion.
John Kessenich [Tue, 3 Jan 2017 00:56:08 +0000 (17:56 -0700)]
PP: Fix issue #426, recover from bad-source macro expansion.

7 years agoSPV: Fix issue #369, don't support gl_NumSamples -> SPIR-V.
John Kessenich [Tue, 3 Jan 2017 00:01:21 +0000 (17:01 -0700)]
SPV: Fix issue #369, don't support gl_NumSamples -> SPIR-V.

7 years agoHLSL: Fix issue #646: map SV_DispatchThreadID -> GlobalInvocationID.
John Kessenich [Mon, 2 Jan 2017 21:59:19 +0000 (14:59 -0700)]
HLSL: Fix issue #646: map SV_DispatchThreadID -> GlobalInvocationID.

7 years agoHLSL: Handle const with no initializer. Fixes issue #651.
John Kessenich [Fri, 30 Dec 2016 23:42:57 +0000 (16:42 -0700)]
HLSL: Handle const with no initializer. Fixes issue #651.

7 years agoHLSL: Support empty {} initializers for arrays and scalars.
John Kessenich [Fri, 30 Dec 2016 22:59:28 +0000 (15:59 -0700)]
HLSL: Support empty {} initializers for arrays and scalars.

7 years agoHLSL default function parameters
steve-lunarg [Sat, 24 Dec 2016 01:56:57 +0000 (18:56 -0700)]
HLSL default function parameters

This PR adds support for default function parameters in the following cases:

1. Simple constants, such as void fn(int x, float myparam = 3)
2. Expressions that can be const folded, such a ... myparam = sin(some_const)
3. Initializer lists that can be const folded, such as ... float2 myparam = {1,2}

New tests are added: hlsl.params.default.frag and hlsl.params.default.err.frag
(for testing error situations, such as ambiguity or non-const-foldable).

In order to avoid sampler method ambiguity, the hlsl better() lambda now
considers sampler matches.  Previously, all sampler types looked identical
since only the basic type of EbtSampler was considered.

7 years agoHLSL: allow scalar type keywords as identifiers, and add half type support.
steve-lunarg [Tue, 27 Dec 2016 01:45:52 +0000 (18:45 -0700)]
HLSL: allow scalar type keywords as identifiers, and add half type support.

HLSL allows type keywords to also be identifiers, so a sequence such as "float half = 3" is
valid, or more bizzarely, something like "float.float = int.uint + bool;"

There are places this is not supported.  E.g, it's permitted for struct members, but not struct
names or functions.  Also, vector or matrix types such as "float3" are not permitted as
identifiers.

This PR adds that support, as well as support for the "half" type.  In production shaders,
this was seen with variables named "half".  The PR attempts to support this without breaking
useful grammar errors such as "; expected" at the end of unterminated statements, so it errs
on that side at the possible expense of failing to accept valid constructs containing a type
keyword identifier.  If others are discovered, they can be added.

Also, half is now accepted as a valid type, alongside the min*float types.

7 years agoHLSL: struct splitting: assignments of hierarchical split types
steve-lunarg [Mon, 19 Dec 2016 22:48:01 +0000 (15:48 -0700)]
HLSL: struct splitting: assignments of hierarchical split types

This commit adds support for copying nested hierarchical types of split
types.  E.g, a struct of a struct containing both user and builtin interstage
IO variables.

When copying split types, if any subtree does NOT contain builtin interstage
IO, we can copy the whole subtree with one assignment, which saves a bunch
of AST verbosity for memberwise copies of that subtree.

7 years agoHLSL: inter-stage structure splitting.
steve-lunarg [Wed, 14 Dec 2016 22:22:25 +0000 (15:22 -0700)]
HLSL: inter-stage structure splitting.

This adds structure splitting, which among other things will enable GS support where input structs
are passed, and thus become input arrays of structs in the GS inputs.  That is a common GS case.

The salient points of this PR are:

* Structure splitting has been changed from "always between stages" to "only into the VS and out of
  the PS".  It had previously happened between stages because it's not legal to pass a struct
  containing a builtin IO variable.

* Structs passed between stages are now split into a struct containing ONLY user types, and a
  collection of loose builtin IO variables, if any.  The user-part is passed as a normal struct
  between stages, which is valid SPIR-V now that the builtin IO is removed.

* Internal to the shader, a sanitized struct (with IO qualifiers removed) is used, so that e.g,
  functions can work unmodified.

* If a builtin IO such as Position occurs in an arrayed struct, for example as an input to a GS,
  the array reference is moved to the split-off loose variable, which is given the array dimension
  itself.

When passing things around inside the shader, such as over a function call, the the original type
is used in a sanitized form that removes the builtIn qualifications and makes them temporaries.
This means internal function calls do not have to change.  However, the type when returned from
the shader will be member-wise copied from the internal sanitized one to the external type.
The sanitized type is used in variable declarations.

When copying split types and unsplit, if a sub-struct contains only user variables, it is copied
as a single entity to avoid more AST verbosity.

Above strategy arrived at with talks with @johnkslang.

This is a big complex change.  I'm inclined to leave it as a WIP until it can get some exposure to
real world cases.

7 years agoMerge pull request #640 from chaoc/modify-shader-ballot
John Kessenich [Thu, 22 Dec 2016 00:40:29 +0000 (17:40 -0700)]
Merge pull request #640 from chaoc/modify-shader-ballot

Modify shader ballot extension by adding OpSubgroupReadInvocationKHR

7 years agoPP: Non-functional: Make a proper class out of the atom <-> string mapping.
John Kessenich [Wed, 21 Dec 2016 20:49:16 +0000 (13:49 -0700)]
PP: Non-functional: Make a proper class out of the atom <-> string mapping.

7 years agoModify shader ballot extension by adding OpSubgroupReadInvocationKHR
chaoc [Tue, 20 Dec 2016 20:44:35 +0000 (12:44 -0800)]
Modify shader ballot extension by adding OpSubgroupReadInvocationKHR

7 years agoPP: Support operator creation with token pasting.
John Kessenich [Wed, 21 Dec 2016 19:32:56 +0000 (12:32 -0700)]
PP: Support operator creation with token pasting.

7 years agoPP: Rationalize names of tokens.
John Kessenich [Wed, 21 Dec 2016 18:55:53 +0000 (11:55 -0700)]
PP: Rationalize names of tokens.

7 years agoMerge pull request #641 from chaoc/passthrough
John Kessenich [Wed, 21 Dec 2016 17:59:07 +0000 (10:59 -0700)]
Merge pull request #641 from chaoc/passthrough

Add support for SPV_NV_geometry_shader_passthrough

7 years agoMerge pull request #644 from hrydgard/override-warning-fixes
John Kessenich [Wed, 21 Dec 2016 17:50:33 +0000 (10:50 -0700)]
Merge pull request #644 from hrydgard/override-warning-fixes

Fix a large number of Clang warnings about inconsistent usage of 'override'

7 years agoFix a large number of warnings about inconsistent usage of 'override' produced by...
Henrik Rydgård [Wed, 21 Dec 2016 11:48:08 +0000 (12:48 +0100)]
Fix a large number of warnings about inconsistent usage of 'override' produced by clang

7 years agoPP: Non-functional: Only use string <-> atom mapping when needed.
John Kessenich [Wed, 21 Dec 2016 04:47:30 +0000 (21:47 -0700)]
PP: Non-functional: Only use string <-> atom mapping when needed.

Also, eliminate the 'atom' field of TPpToken.

Parsing a real 300 line shader, through to making the AST, is about 10% faster.

Memory is slightly reduced (< 1%).

The whole google-test suite, inclusive of all testing overhead, SPIR-V generation,
etc., runs 3% faster.

Since this is a code *simplification* that leads to perf. improvement, I'm not
going to invest too much more in measuring the perf. than this. The PP code is
simply now in a better state to see how to further rationalize/improve it.

7 years agoPP: Non-functional: rationalize TPpToken.
John Kessenich [Wed, 21 Dec 2016 02:42:53 +0000 (19:42 -0700)]
PP: Non-functional: rationalize TPpToken.

Always keep 'token' outside.
Always return the string to upper levels inside.

7 years agoPP: Non-functional: clean up, simplify, completely identical operation.
John Kessenich [Tue, 20 Dec 2016 18:10:09 +0000 (11:10 -0700)]
PP: Non-functional: clean up, simplify, completely identical operation.

7 years agoAdd support for SPV_NV_geometry_shader_passthrough
chaoc [Tue, 20 Dec 2016 21:28:52 +0000 (13:28 -0800)]
Add support for SPV_NV_geometry_shader_passthrough

7 years agoMerge pull request #639 from hrydgard/master
John Kessenich [Tue, 20 Dec 2016 18:02:11 +0000 (11:02 -0700)]
Merge pull request #639 from hrydgard/master

CMake: Add option to make it possible to not build the executables

7 years agoMerge pull request #638 from chaoc/master
John Kessenich [Tue, 20 Dec 2016 18:01:21 +0000 (11:01 -0700)]
Merge pull request #638 from chaoc/master

Add support for SPV_NV_sample_mask_override_coverage

7 years agoPP: Non-functional: Remove custom allocator and related improvements.
John Kessenich [Tue, 20 Dec 2016 04:57:06 +0000 (21:57 -0700)]
PP: Non-functional: Remove custom allocator and related improvements.

Removed the preprocesser memory pool.

Removed extra copies and unnecessary allocations of objects related to the ones
that were using the pool.

Replaced some allocated pointers with objects instead, generally using more
modern techiques. There end up being fewer memory allocations/deletions to get right.

Overall combined effect of all changes is to use slightly less memory and
run slightly faster (< 1% for both, but noticable).

As part of simplifying the code base, this change makes it easier to see
PP symbol tracking, which I suspect has an even bigger run-time simplification
to make.

7 years agoAdd option to make it possible not to build the executables
Henrik Rydgård [Tue, 20 Dec 2016 00:56:00 +0000 (01:56 +0100)]
Add option to make it possible not to build the executables

7 years agoPP: Add missing i64val code.
John Kessenich [Mon, 19 Dec 2016 22:32:04 +0000 (15:32 -0700)]
PP: Add missing i64val code.

Also, checking both 'atom' and 'name' is redundant, and I'm hoping to
eliminate more atom stuff.

7 years agoAdd support for SPV_NV_sample_mask_override_coverage
chaoc [Tue, 20 Dec 2016 00:29:34 +0000 (16:29 -0800)]
Add support for SPV_NV_sample_mask_override_coverage

7 years agoMerge remote-tracking branch 'refs/remotes/KhronosGroup/master'
chaoc [Tue, 20 Dec 2016 00:21:14 +0000 (16:21 -0800)]
Merge remote-tracking branch 'refs/remotes/KhronosGroup/master'

7 years agoBuild: Fix #633, add missing overrides.
John Kessenich [Mon, 19 Dec 2016 21:43:42 +0000 (14:43 -0700)]
Build: Fix #633, add missing overrides.

7 years agoMerge pull request #637 from KhronosGroup/token-paste
John Kessenich [Mon, 19 Dec 2016 21:31:57 +0000 (14:31 -0700)]
Merge pull request #637 from KhronosGroup/token-paste

PP: Implement token pasting for PP identifiers.

7 years agoMerge pull request #624 from steve-lunarg/remapper-strip-removed
John Kessenich [Mon, 19 Dec 2016 21:07:58 +0000 (14:07 -0700)]
Merge pull request #624 from steve-lunarg/remapper-strip-removed

WIP: Remapper: remove debug info for IDs stripped in other passes

7 years agoMerge pull request #635 from steve-lunarg/sample-fix-2
John Kessenich [Mon, 19 Dec 2016 20:30:29 +0000 (13:30 -0700)]
Merge pull request #635 from steve-lunarg/sample-fix-2

HLSL: allow "sample" in expressions.

7 years agoMerge pull request #625 from jbeich/gcc6
John Kessenich [Mon, 19 Dec 2016 20:30:06 +0000 (13:30 -0700)]
Merge pull request #625 from jbeich/gcc6

Unbreak build on FreeBSD with GCC/libstdc++ 6.2.0

7 years agoPP: Implement token pasting for PP identifiers.
John Kessenich [Mon, 19 Dec 2016 16:19:43 +0000 (09:19 -0700)]
PP: Implement token pasting for PP identifiers.

Implement token pasting as per the C++ specification, within the current
style of the PP code.
Non-identifiers (turning 12 ## 10 into the numeral 1210) is not yet covered;
they should be a simple incremental change built on this one.
Addresses issue #255.

7 years agoHLSL: allow "sample" in expressions.
steve-lunarg [Mon, 19 Dec 2016 00:51:14 +0000 (17:51 -0700)]
HLSL: allow "sample" in expressions.

Unlike other qualifiers, HLSL allows "sample" to be either a qualifier keyword or an
identifier (e.g, a variable or function name).

A fix to allow this was made a while ago, but that fix was insufficient when 'sample'
was used in an expression.  The problem was around the initial ambiguity between:

   sample float a; // "sample" is part of a fully specified type
and
   sample.xyz;     // sample is a keyword in a dot expression

Both start the same.  The "sample" was being accepted as a qualifier before enough
further parsing was done to determine we were not a declaration after all.  This
consumed the token, causing it to fail for its real purpose.

Now, when accepting a fully specified type, the token is pushed back onto the stack if
the thing is not a fully specified type.  This leaves it available for subsequent
purposes.

Changed the "hlsl.identifier.sample.frag" test to exercise this situation, distilled
down from a production shaders.

7 years agoFix comment typo.
John Kessenich [Sat, 17 Dec 2016 00:11:18 +0000 (17:11 -0700)]
Fix comment typo.

7 years agoHLSL: More robust handling of bad shader input, catching a few more things.
John Kessenich [Thu, 15 Dec 2016 00:02:32 +0000 (17:02 -0700)]
HLSL: More robust handling of bad shader input, catching a few more things.

7 years agoMerge pull request #629 from null77/fix-unicode
John Kessenich [Wed, 14 Dec 2016 22:15:41 +0000 (15:15 -0700)]
Merge pull request #629 from null77/fix-unicode

Change unicode dash to ASCII.

7 years agoChange unicode dash to ASCII.
Jamie Madill [Wed, 14 Dec 2016 20:48:56 +0000 (15:48 -0500)]
Change unicode dash to ASCII.

This change is helpful for integration with Chromium, which recently
added a compiler option to warn when compiling any source files which
use extended characters. In this case the offending character was a
single unicode dash in a comment.

7 years agoMerge pull request #628 from null77/fix-cast-warn
John Kessenich [Wed, 14 Dec 2016 02:32:01 +0000 (19:32 -0700)]
Merge pull request #628 from null77/fix-cast-warn

Fix size_t to int cast warnings.

7 years agoFix size_t to int cast warnings.
Jamie Madill [Tue, 13 Dec 2016 22:30:58 +0000 (17:30 -0500)]
Fix size_t to int cast warnings.

Several instances in Visual Studio 2015:

warning C4267: 'initializing': conversion from 'size_t' to 'int', possible loss of data

7 years agoFix last commit; EOptionKeepUncalled incorrect enum bug.
John Kessenich [Mon, 12 Dec 2016 18:41:43 +0000 (11:41 -0700)]
Fix last commit; EOptionKeepUncalled incorrect enum bug.

7 years agoRemapper: remove debug info for IDs stripped in other passes
steve-lunarg [Fri, 9 Dec 2016 18:13:23 +0000 (11:13 -0700)]
Remapper: remove debug info for IDs stripped in other passes

If some DCE is performed such as removing dead functions, then even
if we are NOT stripping debug info, we still must remove the debug
opcodes that refer to the now-dead IDs.

Also, this adds a small change to perform no ID remapping if none
is requested, making spirv-remap properly be a no-op if no options
are given.

7 years agoLinker: Eliminate uncalled functions, because they can be ill-defined.
John Kessenich [Sat, 10 Dec 2016 02:22:20 +0000 (19:22 -0700)]
Linker: Eliminate uncalled functions, because they can be ill-defined.

Fixes issue #610. Also provides a testing option to keep uncalled functions.

7 years agoLinker: Track the mangled entry-point name along with the non-mangled one.
John Kessenich [Sat, 10 Dec 2016 00:29:07 +0000 (17:29 -0700)]
Linker: Track the mangled entry-point name along with the non-mangled one.

7 years agoGLSL Linker: Track entry point across compilation units.
John Kessenich [Sat, 10 Dec 2016 00:14:27 +0000 (17:14 -0700)]
GLSL Linker: Track entry point across compilation units.

This wasn't needed until the recent generalization of "main" to "entry point",
so makes some HLSL-specific code be generic now, for GLSL functional correctness.

7 years agoAdd missing header and drop duplicate one
Jan Beich [Fri, 9 Dec 2016 03:32:01 +0000 (03:32 +0000)]
Add missing header and drop duplicate one

In file included from C:/Projects/glslang/glslang/MachineIndependent/glslang.y:59:0:
glslang/MachineIndependent/ParseHelper.h:276:24: error: 'va_list' has not been declared
                        va_list args);
                        ^~~~~~~

7 years agoLinker: Walk the call graph to report an error on missing bodies.
John Kessenich [Fri, 9 Dec 2016 04:01:59 +0000 (21:01 -0700)]
Linker: Walk the call graph to report an error on missing bodies.

7 years agoMerge pull request #621 from steve-lunarg/recursive-flattening
John Kessenich [Thu, 8 Dec 2016 18:18:07 +0000 (11:18 -0700)]
Merge pull request #621 from steve-lunarg/recursive-flattening

HLSL: Recursive composite flattening

7 years agoMerge pull request #619 from steve-lunarg/opcode-specific-promote
John Kessenich [Thu, 8 Dec 2016 18:17:21 +0000 (11:17 -0700)]
Merge pull request #619 from steve-lunarg/opcode-specific-promote

HLSL: opcode specific promotion rules for interlocked ops

7 years agoMerge pull request #620 from fjhenigman/foo
John Kessenich [Thu, 8 Dec 2016 18:15:59 +0000 (11:15 -0700)]
Merge pull request #620 from fjhenigman/foo

GLSL: Always define TShader::~Includer().

7 years agoHLSL: Recursive composite flattening
steve-lunarg [Tue, 29 Nov 2016 00:09:54 +0000 (17:09 -0700)]
HLSL: Recursive composite flattening

This PR implements recursive type flattening.  For example, an array of structs of other structs
can be flattened to individual member variables at the shader interface.

This is sufficient for many purposes, e.g, uniforms containing opaque types, but is not sufficient
for geometry shader arrayed inputs.  That will be handled separately with structure splitting,
 which is not implemented by this PR.  In the meantime, that case is detected and triggers an error.

The recursive flattening extends the following three aspects of single-level flattening:

- Flattening of structures to individual members with names such as "foo[0].samp[1]";

- Turning constant references to the nested composite type into a reference to a particular
  flattened member.

- Shadow copies between arrays of flattened members and the nested composite type.

Previous single-level flattening only flattened at the shader interface, and that is unchanged by this PR.
Internally, shadow copies are, such as if the type is passed to a function.

Also, the reasons for flattening are unchanged.  Uniforms containing opaque types, and interface struct
types are flattened.  (The latter will change with structure splitting).

One existing test changes: hlsl.structin.vert, which did in fact contain a nested composite type to be
flattened.

Two new tests are added: hlsl.structarray.flatten.frag, and hlsl.structarray.flatten.geom (currently
issues an error until type splitting is online).

The process of arriving at the individual member from chained postfix expressions is more complex than
it was with one level.  See large-ish comment above HlslParseContext::flatten() for details.

7 years agoHLSL: opcode specific promotion rules for interlocked ops
steve-lunarg [Tue, 6 Dec 2016 22:50:11 +0000 (15:50 -0700)]
HLSL: opcode specific promotion rules for interlocked ops

PR #577 addresses most but not all of the intrinsic promotion problems.
This PR resolves all known cases in the remainder.

Interlocked ops need special promotion rules because at the time
of function selection, the first argument has not been converted
to a buffer object.  It's just an int or uint, but you don't want
to convert THAT argument, because that implies converting the
buffer object itself.  Rather, you can convert other arguments,
but want to stay in the same "family" of functions.  E.g, if
the first interlocked arg is a uint, use only the uint family,
never the int family, you can convert the other args as you please.

This PR allows making such opcode and arg specific choices by
passing the op and arg to the convertible lambda.  The code in
the new test "hlsl.promote.atomic.frag" would not compile without
this change, but it must compile.

Also, it provides better handling of downconversions (to "worse"
types), which are permitted in HLSL.  The existing method of
selecting upconversions is unchanged, but if that doesn't find
any valid ones, then it will allow downconversions.  In effect
this always uses an upconversion if there is one.

7 years agoGLSL: Always define TShader::~Includer().
Frank Henigman [Tue, 6 Dec 2016 21:35:12 +0000 (16:35 -0500)]
GLSL: Always define TShader::~Includer().

Rather than update the existing ifdef to cover all necessary cases,
get rid of it and always define TShader::~Includer().

7 years agoMerge pull request #615 from dneto0/consistent-location-for-test-files
John Kessenich [Tue, 6 Dec 2016 07:48:11 +0000 (00:48 -0700)]
Merge pull request #615 from dneto0/consistent-location-for-test-files

runtests should refer to test files in current directory

7 years agoGLSL: Allow desktop shaders to call functions from outside main().
John Kessenich [Tue, 6 Dec 2016 04:36:48 +0000 (21:36 -0700)]
GLSL: Allow desktop shaders to call functions from outside main().

Fixes issue #239.

7 years agoMerge pull request #613 from Corillian/hlslpreprocess
John Kessenich [Mon, 5 Dec 2016 17:25:02 +0000 (10:25 -0700)]
Merge pull request #613 from Corillian/hlslpreprocess

Fixed processing #include's when preprocessing HLSL

7 years agoruntests should refer to test files in current directory
David Neto [Mon, 5 Dec 2016 16:30:02 +0000 (11:30 -0500)]
runtests should refer to test files in current directory

Recently added entry point renaming file referred to
test source file hlsl.entry.rename.frag via relative directory.

Change it to be consistent with other tests: assume test
sources are in the current directory.

7 years agoFixed processing #include's when preprocessing HLSL
Keith Newton [Sun, 4 Dec 2016 22:07:10 +0000 (17:07 -0500)]
Fixed processing #include's when preprocessing HLSL

7 years agoDon't print the "Linked stage" message unless the AST is being dumped.
John Kessenich [Sat, 3 Dec 2016 20:58:46 +0000 (13:58 -0700)]
Don't print the "Linked stage" message unless the AST is being dumped.

7 years agoMerge branch 'steve-lunarg-intrinsic-promotion'
John Kessenich [Sat, 3 Dec 2016 20:29:46 +0000 (13:29 -0700)]
Merge branch 'steve-lunarg-intrinsic-promotion'

7 years agoMerge branch 'intrinsic-promotion' of https://github.com/steve-lunarg/glslang into...
John Kessenich [Sat, 3 Dec 2016 20:27:22 +0000 (13:27 -0700)]
Merge branch 'intrinsic-promotion' of https://github.com/steve-lunarg/glslang into steve-lunarg-intrinsic-promotion

7 years agoMerge pull request #604 from steve-lunarg/entry-rename-2
John Kessenich [Fri, 2 Dec 2016 06:47:49 +0000 (23:47 -0700)]
Merge pull request #604 from steve-lunarg/entry-rename-2

HLSL: allow renaming of shader entry point when creating SPIR-V

7 years agoallow renaming of shader entry point when creating SPIR-V
steve-lunarg [Mon, 31 Oct 2016 21:13:43 +0000 (15:13 -0600)]
allow renaming of shader entry point when creating SPIR-V

Use "--source-entrypoint name" on the command line, or the
TShader::setSourceEntryPoint(char*) API.

When the name given to the above interfaces is detected in the
shader source, it will be renamed to the entry point name supplied
to the -e option or the TShader::setEntryPoint() method.

7 years agoFix non-Windows build error.
John Kessenich [Mon, 28 Nov 2016 06:00:14 +0000 (23:00 -0700)]
Fix non-Windows build error.

7 years agoHLSL: Support the constructor idiom "(struct type)0".
John Kessenich [Mon, 28 Nov 2016 05:51:36 +0000 (22:51 -0700)]
HLSL: Support the constructor idiom "(struct type)0".

This highly leverages the previous commit to handle partial initializers.

7 years agoHLSL: Support {...} initializer lists that are too short.
John Kessenich [Mon, 28 Nov 2016 00:39:07 +0000 (17:39 -0700)]
HLSL: Support {...} initializer lists that are too short.

7 years agoFix crash by returning early from finalCheck() if there is no tree to process.
John Kessenich [Mon, 28 Nov 2016 00:32:19 +0000 (17:32 -0700)]
Fix crash by returning early from finalCheck() if there is no tree to process.

7 years agoNon-functional: Change a bunch of 0 to nullptr.
John Kessenich [Mon, 28 Nov 2016 00:30:14 +0000 (17:30 -0700)]
Non-functional: Change a bunch of 0 to nullptr.

7 years agoNon-functional: Fix typos.
John Kessenich [Mon, 28 Nov 2016 00:26:21 +0000 (17:26 -0700)]
Non-functional: Fix typos.

7 years agoWarn on HLSL not finding entry point. Issue #588.
John Kessenich [Sat, 26 Nov 2016 20:45:18 +0000 (13:45 -0700)]
Warn on HLSL not finding entry point. Issue #588.

7 years agoNon-functional: Rename some entry-point variables to entryPoint, not main.
John Kessenich [Sat, 26 Nov 2016 20:31:47 +0000 (13:31 -0700)]
Non-functional: Rename some entry-point variables to entryPoint, not main.

7 years agoAlways correctly terminate main. Issue #588, PR #600.
John Kessenich [Sat, 26 Nov 2016 20:23:20 +0000 (13:23 -0700)]
Always correctly terminate main. Issue #588, PR #600.

7 years agoHLSL: add intrinsic function implicit promotions
steve-lunarg [Wed, 2 Nov 2016 18:42:34 +0000 (12:42 -0600)]
HLSL: add intrinsic function implicit promotions

This PR handles implicit promotions for intrinsics when there is no exact match,
such as for example clamp(int, bool, float).  In this case the int and bool will
be promoted to a float, and the clamp(float, float, float) form used.

These promotions can be mixed with shape conversions, e.g, clamp(int, bool2, float2).

Output conversions are handled either via the existing addOutputArgumentConversion
function, which this PR generalizes to handle either aggregates or unaries, or by
intrinsic decomposition.  If there are methods or intrinsics to be decomposed,
then decomposition is responsible for any output conversions, which turns out to
happen automatically in all current cases.  This can be revisited once inout
conversions are in place.

Some cases of actual ambiguity were fixed in several tests, e.g, spv.register.autoassign.*

Some intrinsics with only uint versions were expanded to signed ints natively, where the
underlying AST and SPIR-V supports that.  E.g, countbits.  This avoids extraneous
conversion nodes.

A new function promoteAggregate is added, and used by findFunction.  This is essentially
a generalization of the "promote 1st or 2nd arg" algorithm in promoteBinary.

The actual selection proceeds in three steps, as described in the comments in
hlslParseContext::findFunction:

1. Attempt an exact match.  If found, use it.
2. If not, obtain the operator from step 1, and promote arguments.
3. Re-select the intrinsic overload from the results of step 2.

7 years agoMerge pull request #601 from BearishSun/master
John Kessenich [Wed, 23 Nov 2016 07:38:32 +0000 (00:38 -0700)]
Merge pull request #601 from BearishSun/master

A way to query "location" qualifier for vertex attributes, using TProgram reflection API

7 years agoMerge pull request #599 from steve-lunarg/gs
John Kessenich [Wed, 23 Nov 2016 07:29:30 +0000 (00:29 -0700)]
Merge pull request #599 from steve-lunarg/gs

HLSL: Add GS support

7 years agoMerge pull request #596 from steve-lunarg/hlsl-intrinsic-parsing
John Kessenich [Wed, 23 Nov 2016 07:19:40 +0000 (00:19 -0700)]
Merge pull request #596 from steve-lunarg/hlsl-intrinsic-parsing

HLSL: use HLSL parser for HLSL intrinsic prototypes, enable int/bool mats

7 years agoMerge pull request #597 from steve-lunarg/sample-keyword-fix
John Kessenich [Wed, 23 Nov 2016 06:33:40 +0000 (23:33 -0700)]
Merge pull request #597 from steve-lunarg/sample-keyword-fix

HLSL: allow "sample" as a valid identifier.

7 years agoAdding a way to retrieve vertex attribute TType using TProgram reflection API (requir...
BearishSun [Tue, 22 Nov 2016 08:53:04 +0000 (09:53 +0100)]
Adding a way to retrieve vertex attribute TType using TProgram reflection API (required in order to query location attributes).

7 years agoWIP: HLSL: Add GS support
steve-lunarg [Thu, 17 Nov 2016 22:04:20 +0000 (15:04 -0700)]
WIP: HLSL: Add GS support

This PR adds:

[maxvertexcount(n)] attributes

point/line/triangle/lineadj/triangleadj qualifiers

PointStream/LineStream/TriangleStream templatized types

Append method on above template types

RestartStrip method on above template types.

8 years agoHLSL: allow "sample" as a valid identifier.
steve-lunarg [Wed, 16 Nov 2016 20:22:11 +0000 (13:22 -0700)]
HLSL: allow "sample" as a valid identifier.

HLSL has keywords for various interpolation modifiers such as "linear",
"centroid", "sample", etc.  Of these, "sample" appears to be special,
as it is also accepted as an identifier string, where the others are not.

This PR adds this ability, so the construct "int sample = 42;" no longer
produces a compilation error.

New test = hlsl.identifier.sample.frag

8 years agoHLSL: use HLSL parser to parse HLSL intrinsic prototypes, enable int/bool mats
steve-lunarg [Tue, 15 Nov 2016 17:11:04 +0000 (10:11 -0700)]
HLSL: use HLSL parser to parse HLSL intrinsic prototypes, enable int/bool mats

This PR adds a CreateParseContext() fn analogous to CreateBuiltInParseables(),
to create a language specific built in parser.  (This code was present before
but not encapsualted in a fn).  This can now be used to create a source language
specific parser for builtins.

Along with this, the code creating HLSL intrinsic prototypes can now produce
them in HLSL syntax, rather than GLSL syntax.  This relaxes certain prior
restrictions at the parser level.  Lower layers (e.g, SPIR-V) may still have
such restrictions, such as around Nx1 matrices: this code does not impact
that.

This PR also fleshes out matrix types for bools and ints, both of which were
partially in place before.  This was easier than maintaining the restrictions
in the HLSL prototype generator to avoid creating protoypes with those types.

Many tests change because the result type from intrinsics moves from "global"
to "temp".

Several new tests are added for the new types.