Fix entry_point() crash on generic compute shaders with integer generic parameters

[Copilot]

module.entry_point() throws an untranslatable nanobind exception on generic compute shaders with integer generic parameters (e.g., <int N>). The Slang API's getLayout() cannot be called on unspecialized generic entry points—it throws an internal exception that bypasses nanobind's exception translation.

module = device.load_module('shader.slang')
# Crashes: "exception could not be translated!"
entry = module.entry_point("my_compute")  # where my_compute<int N>(...)

Changes

• SlangEntryPoint::init(): Check getSpecializationParamCount() > 0 before calling getLayout(). For unspecialized generics, use the descriptor name and ShaderStage::none as temporaries—both are resolved correctly when specialize() is later called.
• SlangModule::entry_points(): Replace getLayout()->getEntryPointByIndex(0)->getName() with getFunctionReflection()->getName(). The latter is available directly on IEntryPoint and works for generic entry points without requiring layout computation.
• Tests: Add integer-generic entry point shader and two tests covering entry_point() and entry_points with <int N> generics.

Original prompt

---

This section details on the original issue you should resolve

<issue_title>module.entry_point() raises on generic compute shader with generic parameters</issue_title>
<issue_description>Invoking module.entry_point() on a generic compute shader that takes one or more generic parameters raises a nanobind exception.

Interestingly, accessing module.entry_points prior to calling entry_point() works around the issue, even though doing so raises itself.

Repro shader and script:

// Generic entry point with generic struct parameter

struct MyData<int N>
{
    float value;
}

[shader("compute")]
[numthreads(8, 1, 1)]
void my_compute<int N>(
    uint3 tid: SV_DispatchThreadID,
    RWStructuredBuffer<MyData<N>> output
)
{
    MyData<N> data;
    data.value = float(tid.x);
    output[tid.x] = data;
}

import slangpy as spy
from pathlib import Path

device = spy.Device(
    compiler_options={
        "include_paths": [spy.SHADER_PATH, Path(__file__).parent.absolute()]
    }
)

print(f"Using device: {device.info.adapter_name}\n")

# Don't use workaround; expect failure
print("1: Call entry_point() directly")
try:
    module = device.load_module('entrypoint.slang')
    entry = module.entry_point("my_compute")
    print("Result: SUCCESS (issue not reproduced)\n")
except Exception as e:
    print(f"Result: FAILED with {e} (issue reproduced)\n")

# Use workaround; expect success
print("2: Call entry_point() after accessing entry_points")
try:
    module = device.load_module('entrypoint.slang')
    try:
        _ = module.entry_points
    except:
        pass
    entry = module.entry_point("my_compute")
    specialized = entry.specialize([spy.SpecializationArg.from_expr("4")])
    program = device.link_program(modules=[module], entry_points=[specialized])
    kernel = device.create_compute_kernel(program)
    print("Result: SUCCESS (workaround succeeded)\n")
except Exception as e:
    print(f"Result: FAILED - {e}\n")

Current output:

Using device: NVIDIA RTX PRO 6000 Blackwell Workstation Edition

1: Call entry_point() directly
Result: FAILED with nanobind::detail::nb_func_error_except(): exception could not be translated! (issue reproduced)

2: Call entry_point() after accessing entry_points
Result: SUCCESS (workaround succeeded)
```</issue_description>

## Comments on the Issue (you are @copilot in this section)

<comments>
</comments>

• Fixes module.entry_point() raises on generic compute shader with generic parameters #788

---

💡 You can make Copilot smarter by setting up custom instructions, customizing its development environment and configuring Model Context Protocol (MCP) servers. Learn more Copilot coding agent tips in the docs.