How to disable name mangling?

[st0rmbtw]

I have a compute shader with a function with uniform parameters like this:

[shader("compute")]
[numthreads(32, 1, 1)]
void CSComputeLightVertical(
    uint3 thread_id: SV_DispatchThreadID,
    uniform uint2 uniform_min,
    uniform uint2 uniform_max
) {
    ...
}

I want to compile it to HLSL. I'm expecting it to compile to something like this:

...

cbuffer UniformBuffer : register(b0)
{
    uint2 uniform_min;
    uint2 uniform_max;
}

[numthreads(32, 1, 1)]
void CSComputeLightVertical(uint3 thread_id_0 : SV_DispatchThreadID)
{
}

...

But instead, I get this:

struct EntryPointParams_0
{
    uint2 uniform_min_0;
    uint2 uniform_max_0;
};

cbuffer entryPointParams_0 : register(b0)
{
    EntryPointParams_0 entryPointParams_0;
}

[numthreads(32, 1, 1)]
void CSComputeLightVertical(uint3 thread_id_0 : SV_DispatchThreadID)
{
}

I need the uniform_* fields not to be mangled. I don't care about other fields in other structs.

And also, why does it declare a separate EntryPointParams_0 struct, if it can place the fields in the cbuffer, like this:

cbuffer entryPointParams_0 : register(b0)
{
    uint2 uniform_min_0;
    uint2 uniform_max_0;
}

[expipiplus1]

we do have a flag no-mangle which instructs the compiler to do as little name mangling as possible.

Also, when querying these field names with the reflection api, the original names are preserved.

The fact that it creates a wrapping struct could be considered an implementation detail as the layout will be the same with or without it.

Please let me know if this flag works for you :)

[st0rmbtw]

we do have a flag no-mangle which instructs the compiler to do as little name mangling as possible.

I'm using the C++ API, not the command line utility. There is the NoMangle compiler option, but enabling it doesn't seem to change anything.

It would be nice to have an attribute, no_mangle for example, so you could disable mangling for specific names. Something like this:

cbuffer UniformBuffer : register(0b) {
	[no_mangle] float2 position;
}

The fact that it creates a wrapping struct could be considered an implementation detail as the layout will be the same with or without it.

Is there a way to disable this behavior? I really want the fields to be in the original struct, not in the wrapping one. Because the graphics library that I use, identifies uniforms (push constants) by their name, and it can't find them if they are in a separate struct.

[expipiplus1]

I don't believe that there is such an option currently. I think it might be best to open a feature request issue and we can go from there, thanks!

[st0rmbtw]

I think it might be best to open a feature request issue and we can go from there, thanks!

Created the issue: #8391