Without a geometry shader to specify, the pixel shader's gl_Layer is always 0.

In order to render a cube (which has 6 layers), from C++ you will set one layer at a time and render over and over again.

When a geometry shader is present, the pixel shader will only produce output for that layer. If you are rendering a cubemap, you can emit 6 triangles by duplicating the original one; and assign a different layer for each.

Mind you, doing this isn't a good idea because geometry shaders are freakishly slow to the point they're nearly useless.

You can use extensions like GL_AMD_vertex_shader_layer to set the gl_Layer from the vertex shader, or GL_NV_geometry_shader_passthrough to use a geometry shader that only emits one triangle with the right gl_Layer and doesn't have the performance hit of a real GS.

If using any of these 2 extensions, if you want an object to render in all 6 cube faces; you will still have to call glDrawElements 6 times (or use instancing) and set the gl_Layer for each; as the pixel shader is ran only once.