Added support for the 'layout' attribute

[BearishSun]

I've added support for the 'layout' attribute that allows the user to specify image layout format such as rgba8 using a non-standard [layout(rgba8)] attribute, as we discussed yesterday.

This functionality is only enabled when XSC_ENABLE_LANGUAGE_EXT is enabled through CMake AND when the necessary extension is enabled through input options/CLI/GUI.

I've added the -E<TYPE> CLI option which works similar to -W<TYPE>, and allows specific (or all) extensions to be enabled. Since I know you are working on an extension of your own I assume this is preferred than having separate options for each extension.

The core additions are:

• ImageLayoutFormat enumeration containing all image layout formats and a mapping to GLSL keywords. This is available even when extensions are disabled, since it is part of standard GLSL, and you might want to use this functionality for the input-option based solution to the image layout format problem (as discussed yesterday).
• New layout attribute and a mapping of its identifiers to ImageLayoutFormat in HLSLAnalyzer. Resulting image layout is stored on relevant BufferTypeDenoters.
• Change to GLSLGenerator::WriteBufferDeclTexture so that image layout stored on BufferTypeDenoter is used if available. If one is not available or extension is disabled the system will try to deduce one automatically from base buffer type (similar how it worked so far). Finally if format cannot be deduced the buffer will be marked as writeonly, since formats are only needed for reads.

I'll probably do a secondary PR in which I track which buffers are actually being written to. In that case we can always avoid writing out the layout format, and since most RW buffers are only used for writes, this should allow the user to avoid attributes in majority of situations.

[BearishSun]

Added another commit in order to avoid writing image layout format qualifiers if images aren't used for any reads (i.e. what I wrote in the last paragraph above).

[LukasBanana]

I just looked at it for a moment, but it looks good so far. The idea of using -E<TYPE> for language extensions is great 👍 , but I think I should then change the -E ENTRY command to -e ENTRY, let's see. Also making the enum ImageLayoutFormat always available is sensible.

[BearishSun]

Ah indeed, I missed the fact that "-E" was already in use, something should get renamed.

[LukasBanana]

Yes, but I appreciate the upper case letters for commands with integrated types like -W, -S, etc. I'll take care of it. Maybe I find another suitable name for the extensions.

[BearishSun]

Thought you might find it useful :)

I amended the last commit, it was incorrectly applying writeonly to non RW buffers.

[LukasBanana]

I renamed your -E language extension command to -X for eXtension.

I also made the extensions member and its enum in ShaderInput always enabled, but it is ignored if XSC_ENABLE_LANGUAGE_EXT is not defined. In this way, the client programmer doesn't need to define XSC_ENABLE_LANGUAGE_EXT for it's own projects (only to bulid the compiler). In the XscDebugger the respective PropertyGrid entry is then disbaled, but visible so that you always see what is build into the ccompiler and what is not.