Fix reflection serialization for D3D11_SHADER_TYPE_DESC

[killerdevildog]

Fix reflection serialization for D3D11_SHADER_TYPE_DESC

Summary

This PR fixes a critical bug in the shader reflection serialization that prevented bgfx (and potentially other users of d3d4linux) from extracting uniform/constant buffer information from compiled HLSL shaders.

The Problem

When bgfx compiles shaders using d3d4linux, it needs to extract reflection data to determine uniform names, types, and sizes. The reflection API provides this through:

ID3D11ShaderReflectionVariable* var = cbuffer->GetVariableByIndex(i);
ID3D11ShaderReflectionType* type = var->GetType();
D3D11_SHADER_TYPE_DESC type_desc;
type->GetDesc(&type_desc);

However, d3d4linux was not serializing the D3D11_SHADER_TYPE_DESC data across the Wine IPC boundary. This meant that while variables were being enumerated, their type information was garbage, causing:

• Uniforms to have incorrect types (e.g., mat4 appearing as unknown)
• Uniform sizes to be wrong or zero
• Shaders to fail at runtime due to incorrect constant buffer layouts

The Fix

Wine side (d3d4linux.cpp)

Added serialization of the type descriptor after each variable:

// After writing variable descriptor...
ID3D11ShaderReflectionType *type = variable->GetType();
D3D11_SHADER_TYPE_DESC type_desc;
type->GetDesc(&type_desc);
p.write_raw(&type_desc, sizeof(type_desc));
p.write_string(type_desc.Name ? type_desc.Name : "");

Client side (d3d4linux_impl.h)

Added deserialization to populate the m_type member:

// After reading variable descriptor...
p.read_raw(&var.m_type.m_desc, sizeof(var.m_type.m_desc));
var.m_type.m_name = p.read_string();

Min Precision Support (d3d4linux.h)

Added the D3D_MIN_PRECISION enum which was missing from the header. This enum is required by the D3D11_SHADER_TYPE_DESC structure to specify minimum precision requirements for shader variables (e.g., D3D_MIN_PRECISION_DEFAULT, D3D_MIN_PRECISION_FLOAT_16, etc.):

enum D3D_MIN_PRECISION
{
    D3D_MIN_PRECISION_DEFAULT   = 0,
    D3D_MIN_PRECISION_FLOAT_16  = 1,
    D3D_MIN_PRECISION_FLOAT_2_8 = 2,
    D3D_MIN_PRECISION_RESERVED  = 3,
    D3D_MIN_PRECISION_SINT_16   = 4,
    D3D_MIN_PRECISION_UINT_16   = 5,
    D3D_MIN_PRECISION_ANY_16    = 0xf0,
    D3D_MIN_PRECISION_ANY_10    = 0xf1,
};

Without this enum, the MinPrecision field in D3D11_SHADER_TYPE_DESC would cause compilation errors or undefined behavior.

Testing

Tested with bgfx shader compiler (shaderc) on Linux:

1. Build: Compiled shaderc with d3d4linux integration
2. Compile: Built cubes example shaders for dx11 profile

   shaderc -f vs_cubes.sc -o vs_cubes.bin --type vertex --platform windows -p vs_5_0
   shaderc -f fs_cubes.sc -o fs_cubes.bin --type fragment --platform windows -p ps_5_0
   

3. Verify: Confirmed u_modelViewProj uniform extracted correctly:
   • Type: D3D_SVT_FLOAT (matrix)
   • Rows: 4, Columns: 4
   • Size: 64 bytes
4. Runtime: Ran cubes example in Wine+DXVK - renders correctly with proper transforms

Before/After

Before: Uniforms had garbage type data, causing runtime failures

Uniform: u_modelViewProj
  Type: 0 (unknown)
  Size: 0

After: Uniforms correctly identified

Uniform: u_modelViewProj  
  Type: D3D_SVT_FLOAT (matrix 4x4)
  Size: 64

Compatibility

This change maintains backward compatibility with the IPC protocol - it only adds data that was previously missing. The ID3D11ShaderReflectionType stub class already had m_desc and m_name members; they just weren't being populated.

Related to bkaradzic/bgfx#3578