Loop getting optimized out in mesh shader

[Dynamitos]

When declaring a simple loop to output data from a mesh shader, the loop completely disappears when outputting glsl code (probably also for other targets, but I can't read spriv-asm and nothing is shown on screen).

Example modified from the test cases:

const static float2 positions[3] = {
  float2(0.0, -0.5),
  float2(0.5, 0.5),
  float2(-0.5, 0.5)
};

const static float3 colors[3] = {
  float3(1.0, 1.0, 0.0),
  float3(0.0, 1.0, 1.0),
  float3(1.0, 0.0, 1.0)
};
struct Vertex
{
  float4 pos : SV_Position;
  float3 color : Color;
  int index : Index;
};

const static uint MAX_VERTS = 12;
const static uint MAX_PRIMS = 4;

[outputtopology("triangle")]
[shader("mesh")]
[numthreads(3, 1, 1)]
void meshMain(
    in uint tig : SV_GroupIndex,
    out Vertices<Vertex, MAX_VERTS> verts,
    out Indices<uint3, MAX_PRIMS> triangles)
{
    const uint numVertices = 12;
    const uint numPrimitives = 4;
    SetMeshOutputCounts(numVertices, numPrimitives);

    for(uint i = tig; i < numVertices; ++i)
    {
        const int tri = i / 3;
        verts[i] = {float4(positions[i % 3], 0, 1), colors[i % 3], tri};
    }

    for(uint i = tig; i < numPrimitives; ++i)
    {
        triangles[i] = i * 3 + uint3(0,1,2);
    }
}

To test, i compiled this on the commandline with: slangc.exe test.slang -profile sm_6_6 -target glsl -entry meshMain -capability spirv_1_5 -o test.glsl

#version 450
#extension GL_EXT_mesh_shader : require
layout(row_major) uniform;
layout(row_major) buffer;

#line 27 0
layout(local_size_x = 3, local_size_y = 1, local_size_z = 1) in;
layout(max_vertices = 12) out;
layout(max_primitives = 4) out;
layout(triangles) out;
void main()
{

    SetMeshOutputsEXT(12U, 4U);

#line 46
    return;
}

[csyonghe]

@expipiplus1 This should be fixed in tandem with #2804.
The Vertices type should not be semantically an out parameter, and its subscript operator should map to a dedicated instruction that will be treated by the compiler as an external memory object and therefore DCE shouldn't apply to any read/writes on that location.

[natduca]

@csyonghe looks like #2804 is put for Q3 , should we move 2804 into Q1, too?

[csyonghe]

Moved #2804 to Q1.