XCode 15 includes a completely new linker implementation.
Hence some flags are deprecated.
The default is now error so we can simply remove the linker option.
* ci: Test macos-12 and macos-13 instead of macos-11 and macos-12
Developer survey shows little to no users still use MacOS 11 or earlier.
* Remove pointless comment about travis
* Fix minor indent issue
NDK 23.2.8568313 was removed from the latest runner
Hardcoding the NDK versions like before is asking for a CI
breakage.
Use ANDROID_NDK_HOME environment variable which is set by the
runner
glslang representing literal constants with double precision, so 1.0e40 and 1.0e-50 are normal values.
Shader1:
precision highp float;
out vec4 my_FragColor;
void main()
{
// Out-of-range floats should overflow to infinity
// GLSL ES 3.00.6 section 4.1.4 Floats:
// "If the value of the floating point number is too large (small) to be stored as a single precision value, it is converted to positive (negative) infinity"
float correct = isinf(1.0e40) ? 1.0 : 0.0;
my_FragColor = vec4(0.0, correct, 0.0, 1.0);
}
The expected ouput result of this test is vec4(0.0, 1.0, 0.0, 1.0),
but it's vec4(0.0,0.0,0.0,1.0).Because the return value of isInf is
false.
precision highp float;
out vec4 my_FragColor;
void main()
{
// GLSL ES 3.00.6 section 4.1.4 Floats:
// "A value with a magnitude too small to be represented as a mantissa and exponent is converted to zero."
// 1.0e-50 is small enough that it can't even be stored as subnormal.
float correct = (1.0e-50 == 0.0) ? 1.0 : 0.0;
my_FragColor = vec4(0.0, correct, 0.0, 1.0);
}
The expected ouput result of this test is vec4(0.0, 1.0, 0.0, 1.0),
but it's vec4(0.0,0.0,0.0,1.0).
For f32 and f16 type, when the literal constant out of range of the f32
and f16 number, the value should overflow or underflow to inf or zero.
glcts test item
KHR-GLES3.number_parsing.float_out_of_range_as_infinity
Only the headers that are part of glslang's public interface are
installed. Previously, all of its headers were installed, which exposed
a lot of internal implementation details and made it difficult to
maintain backward compatiblity. This reduces the API surface somewhat
and will make it easier to maintain API and ABI compatibility.
The dependency was only because of the SpvOptions struct which is used
in both, but really is part of the glslang public interface and should
be in the public GlslangToSpv.h header.
GlslangToSpv.h only declares functions with opaque references to to
TIntermediate, for which a simple "class TIntermediate;" declaration is
adequate. This means that Include/intermediate.h no longer needs to be
installed as part of glslang's public interface.
The distutils package was removed in Python 3.12, however its only
usage in this script can easily be replaced with functionality available
in the builtin os package in Python 3.2 and later.
Fixes#3393
The debug options passed down from the public ShConstruct* functions to
internal code is unused. This change removes the internal debugOptions
fields and attempts to make it more obvious these fields are not used.
This change does not change the public-facing interface.
This change also adds the -Wshorten-64-to-32 warning to the StandAlone
build in order to avoid unwanted 64-to-32 bit conversions in the future.
Closes#3348.
The Windows tests job in Github Actions was incorrectly reporting
'success' on failed runs. This change ensures a single command is run in
each test 'step' to help remove any guesswork w.r.t. the shell behavior
on each platform.
This change also takes the oppportunity to make the formatting more
consistent in continuous_integration.yml.
1. Pull OpDebugFunction, OpDebugScope and OpDebugVariable for params out
of makeFunctionEntry.
2. Put above in a separate function called setupDebugFunctionEntry,
which also accept line number and set it correctly in builder.
3. Call setupDebugFunctionEntry in makeFunction. Also special case
handle entry function since it's created ealier elsewhere.
Currently no debug info is emitted for buffer reference types, which
resulted in the SPIR-V backend code asserting when trying to emit the
debug info for struct member that had such a type. Instead, the code now
skips such struct members. Full debug info for buffer references may
require forward references in the debug info instructions, which is
currently prohibited by the spec.
SPIR-V 1.6 added the LocalSizeId execution mode that allows using
spec constants for setting the work-group size, however it does not
deprecate the LocalSize mode. This change causes the LocalSizeId mode to
only be used when at least one of the workgroup size is actually
specified with a spec constant.
Fixes#3200
Move the parameter verifictation to a centralized place where all the builtins
are verified for correctness.
Add verification for the new builtins with version and extension check
These builtins are supported on GLSL since version 130 and GLES since
version 300.