//===----------------------------------------------------------------------===//
//
// Part of CUDA Experimental in CUDA C++ Core Libraries,
// under the Apache License v2.0 with LLVM Exceptions.
// See https://llvm.org/LICENSE.txt for license information.
// SPDX-License-Identifier: Apache-2.0 WITH LLVM-exception
// SPDX-FileCopyrightText: Copyright (c) 2025 NVIDIA CORPORATION & AFFILIATES.
//
//===----------------------------------------------------------------------===//

#include <cuda/__driver/driver_api.h>
#include <cuda/devices>
#include <cuda/std/cstddef>
#include <cuda/std/optional>
#include <cuda/std/type_traits>
#include <cuda/std/utility>

#include <cuda/experimental/__utility/ensure_current_device.cuh>
#include <cuda/experimental/kernel.cuh>
#include <cuda/experimental/library.cuh>

#include <testing.cuh>

// extern "C" __constant__ int const_data;
//
// extern "C" __device__ int global_data;
//
// extern "C" __managed__ int managed_data;
//
// extern "C" __global__ void kernel(int* array, int n)
// {
//   __shared__ int shared[32];
//   int tid = blockDim.x * blockIdx.x + threadIdx.x;
//   if (tid < n)
//   {
//     shared[threadIdx.x] = array[tid];
//     __syncthreads();
//     array[tid] = shared[threadIdx.x + 1 % 32] + const_data;
//   }
// }

constexpr char library_src[] = R"(
//
// Generated by NVIDIA NVVM Compiler
//
// Compiler Build ID: CL-32267302
// Cuda compilation tools, release 12.0, V12.0.140
// Based on NVVM 7.0.1
//

.version 8.0
.target sm_75
.address_size 64

	// .globl	kernel
.const .align 4 .u32 const_data;
.global .align 4 .u32 global_data;
.global .attribute(.managed) .align 4 .u32 managed_data;
// _ZZ6kernelE6shared has been demoted

.visible .entry kernel(
	.param .u64 kernel_param_0,
	.param .u32 kernel_param_1
)
{
	.reg .pred 	%p<2>;
	.reg .b32 	%r<13>;
	.reg .b64 	%rd<5>;
	// demoted variable
	.shared .align 4 .b8 _ZZ6kernelE6shared[128];

	ld.param.u64 	%rd1, [kernel_param_0];
	ld.param.u32 	%r3, [kernel_param_1];
	mov.u32 	%r4, %ntid.x;
	mov.u32 	%r5, %ctaid.x;
	mov.u32 	%r1, %tid.x;
	mad.lo.s32 	%r2, %r4, %r5, %r1;
	setp.ge.s32 	%p1, %r2, %r3;
	@%p1 bra 	$L__BB0_2;

	cvta.to.global.u64 	%rd2, %rd1;
	mul.wide.s32 	%rd3, %r2, 4;
	add.s64 	%rd4, %rd2, %rd3;
	ld.global.u32 	%r6, [%rd4];
	shl.b32 	%r7, %r1, 2;
	mov.u32 	%r8, _ZZ6kernelE6shared;
	add.s32 	%r9, %r8, %r7;
	st.shared.u32 	[%r9], %r6;
	bar.sync 	0;
	ld.const.u32 	%r10, [const_data];
	ld.shared.u32 	%r11, [%r9+4];
	add.s32 	%r12, %r10, %r11;
	st.global.u32 	[%rd4], %r12;

$L__BB0_2:
	ret;

}
)";

C2H_CCCLRT_TEST("Library reference", "[library_ref]")
{
  constexpr char kernel_name[]         = "kernel";
  constexpr char global_symbol_name[]  = "global_data";
  constexpr char const_symbol_name[]   = "const_data";
  constexpr char managed_symbol_name[] = "managed_data";

  CUlibrary lib1 = ::cuda::__driver::__libraryLoadData(library_src, nullptr, nullptr, 0, nullptr, nullptr, 0);
  CUlibrary lib2 = ::cuda::__driver::__libraryLoadData(library_src, nullptr, nullptr, 0, nullptr, nullptr, 0);

  const cuda::device_ref device{0};

  // Types
  {
    STATIC_REQUIRE(cuda::std::is_same_v<typename cudax::library_ref::value_type, CUlibrary>);
  }

  // Default constructor
  {
    STATIC_REQUIRE(!cuda::std::is_default_constructible_v<cudax::library_ref>);
  }

  // Constructor from library handle
  {
    STATIC_REQUIRE(cuda::std::is_constructible_v<cudax::library_ref, CUlibrary>);
    STATIC_REQUIRE(cuda::std::is_convertible_v<CUlibrary, cudax::library_ref>);

    cudax::library_ref lib_ref{lib1};
    REQUIRE(lib1 == lib_ref.get());
  }

  // Copy constructor
  {
    STATIC_REQUIRE(cuda::std::is_trivially_copy_constructible_v<cudax::library_ref>);

    cudax::library_ref lib_ref1{lib1};
    REQUIRE(lib1 == lib_ref1.get());

    cudax::library_ref lib_ref2{lib1};
    REQUIRE(lib1 == lib_ref2.get());
    REQUIRE(lib_ref1.get() == lib_ref2.get());
  }

  // Has kernel
  {
    STATIC_REQUIRE(
      cuda::std::is_same_v<decltype(cuda::std::declval<cudax::library_ref>().has_kernel(kernel_name)), bool>);

    cudax::library_ref lib_ref{lib1};
    REQUIRE(lib_ref.has_kernel(kernel_name));
    REQUIRE(!lib_ref.has_kernel("non_existent_kernel"));
  }

  // Get kernel
  {
    STATIC_REQUIRE(
      cuda::std::is_same_v<decltype(cuda::std::declval<cudax::library_ref>().kernel<void(int*, int)>(kernel_name)),
                           cudax::kernel_ref<void(int*, int)>>);

    cudax::library_ref lib_ref{lib1};
    auto kernel = lib_ref.kernel<void(int*, int)>(kernel_name);

    CUkernel kernel_handle;
    REQUIRE_CUDART(::cuda::__driver::__libraryGetKernelNoThrow(kernel_handle, lib1, kernel_name));
    REQUIRE(kernel.get() == kernel_handle);
  }

  // Has global symbol
  {
    STATIC_REQUIRE(
      cuda::std::is_same_v<decltype(cuda::std::declval<cudax::library_ref>().has_global(global_symbol_name, device)),
                           bool>);

    cudax::library_ref lib_ref{lib1};
    REQUIRE(lib_ref.has_global(global_symbol_name, device));
    REQUIRE(lib_ref.has_global(const_symbol_name, device));
    REQUIRE(!lib_ref.has_global("non_existent_global", device));
  }

  // Get global symbol
  {
    STATIC_REQUIRE(
      cuda::std::is_same_v<decltype(cuda::std::declval<cudax::library_ref>().global(global_symbol_name, device)),
                           cudax::library_symbol_info>);

    cudax::library_ref lib_ref{lib1};

    // Test global_symbol_name
    {
      auto global_sym = lib_ref.global(global_symbol_name, device);

      cuda::__ensure_current_context context_guard{device};

      CUdeviceptr global_symbol_ptr;
      cuda::std::size_t global_symbol_size;
      REQUIRE_CUDART(
        ::cuda::__driver::__libraryGetGlobalNoThrow(global_symbol_ptr, global_symbol_size, lib1, global_symbol_name));

      REQUIRE(reinterpret_cast<CUdeviceptr>(global_sym.ptr) == global_symbol_ptr);
      REQUIRE(global_sym.size == global_symbol_size);
      REQUIRE(global_sym.size == sizeof(int));
    }

    // Test const_symbol_name
    {
      auto const_sym = lib_ref.global(const_symbol_name, device);

      cuda::__ensure_current_context context_guard{device};

      CUdeviceptr const_symbol_ptr;
      cuda::std::size_t const_symbol_size;
      REQUIRE_CUDART(
        ::cuda::__driver::__libraryGetGlobalNoThrow(const_symbol_ptr, const_symbol_size, lib1, const_symbol_name));

      REQUIRE(reinterpret_cast<CUdeviceptr>(const_sym.ptr) == const_symbol_ptr);
      REQUIRE(const_sym.size == const_symbol_size);
      REQUIRE(const_sym.size == sizeof(int));
    }
  }

  // Has managed symbol
  {
    STATIC_REQUIRE(
      cuda::std::is_same_v<decltype(cuda::std::declval<cudax::library_ref>().has_managed(managed_symbol_name)), bool>);

    cudax::library_ref lib_ref{lib1};
    REQUIRE(lib_ref.has_managed(managed_symbol_name));
    REQUIRE(!lib_ref.has_managed("non_existent_managed"));
  }

  // Get managed symbol
  {
    STATIC_REQUIRE(cuda::std::is_same_v<decltype(cuda::std::declval<cudax::library_ref>().managed(managed_symbol_name)),
                                        cudax::library_symbol_info>);

    cudax::library_ref lib_ref{lib1};
    auto managed_sym = lib_ref.managed(managed_symbol_name);

    CUdeviceptr managed_symbol_ptr;
    cuda::std::size_t managed_symbol_size;
    REQUIRE_CUDART(
      ::cuda::__driver::__libraryGetManagedNoThrow(managed_symbol_ptr, managed_symbol_size, lib1, managed_symbol_name));

    REQUIRE(reinterpret_cast<CUdeviceptr>(managed_sym.ptr) == managed_symbol_ptr);
    REQUIRE(managed_sym.size == managed_symbol_size);
    REQUIRE(managed_sym.size == sizeof(int));
  }

  // Get handle
  {
    STATIC_REQUIRE(cuda::std::is_same_v<decltype(cuda::std::declval<cudax::library_ref>().get()), CUlibrary>);

    cudax::library_ref lib_ref{lib1};
    REQUIRE(lib1 == lib_ref.get());
  }

  // Equality/Inequality comparison
  {
    cudax::library_ref lib_ref1{lib1};
    cudax::library_ref lib_ref2{lib2};

    REQUIRE(lib_ref1 == lib_ref1);
    REQUIRE(lib_ref1 != lib_ref2);
  }

  REQUIRE_CUDART(::cuda::__driver::__libraryUnloadNoThrow(lib1));
  REQUIRE_CUDART(::cuda::__driver::__libraryUnloadNoThrow(lib2));
}