Throw Exception When Data Layout Will Encounter Alignment Differences With DXC '-fvk-use-dx-layout'

[galibzon]

AZSLc is used to transpile AZSL code into HLSL code, during this process AZSLc can generate reflection data along with size and offset information for all structs, classes and SRGs. AZSLc follows DX12 rules to calculate size and offset of the reflected data.

The idea is that eventually when compiling the HLSL code with DXC, the DXC-calculated sizes and offsets match those calculated by AZSLc. This is true when compiling HLSL code to DX12 byte code.

Natively the layout rules of SPIRV (Vulkan) are different than DX12, hence We request DXC to use DX12 layout rules when compiling for SPIRV with the option '-fvk-use-dx-layout'. The problem is that there are some cases where the expected layout don't match causing crashes or unexpected behavior at runtime.

The idea is to add a feature that checks for those error cases, treat them as errors and provide a message with a data padding solution. Additionally, provide a command line option '--no-alignment-validation' to avoid check for these kind of errors.

The most common error cases are:

1. A 'float' or 'float2' variable preceded by matrices or structs that end in matrices of type:
   float2x2, float3x2, float4x2
   In such case the 'float' or 'float2' should be be pre-padded by a 'float3' variable.

2. A 'float' type variable is preceded by a matrix or struct that ends in matrix of type:
   float2x3, float3x3, float4x3
   In such case the 'float' should be pre-padded by a 'float2' variables.

When AZSLc finds cases in the code that will have layout differences between dx12 and vulkan it will print this message:

tests\Emission\AsError\matRC_padding.azsl(,)
: IR error #131: Detected potential alignment issues related with DXC
flag '-fvk-use-dx-layout'.
Alternatively you can use the option '--no-alignment-validation' to void this error and compile as is:

A 'float3' variable should be added before the variable 'm_f3' in
'struct /A/SD2/m_f3' at Line number 21 of
'D:\o3de-azslc\tests\Emission\AsError\matRC_padding.azsl'
A 'float3' variable should be added before the variable 'm_f5' in
'struct /A/SD2/m_f5' at Line number 23 of
'D:\o3de-azslc\tests\Emission\AsError\matRC_padding.azsl'

---

NOTE: Allegedly the most recent versions of DXC, in particular v1.6.2106+, fixes the issues when using '-fvk-use-dx-layout'.
This needs to be validated. See: microsoft/DirectXShaderCompiler#3945