TheRock/rocm-libraries/projects/rocrand/test/test_rocrand_kernel_sobol64.cpp

// Copyright (c) 2021-2025 Advanced Micro Devices, Inc. All rights reserved.
//
// Permission is hereby granted, free of charge, to any person obtaining a copy
// of this software and associated documentation files (the "Software"), to deal
// in the Software without restriction, including without limitation the rights
// to use, copy, modify, merge, publish, distribute, sublicense, and/or sell
// copies of the Software, and to permit persons to whom the Software is
// furnished to do so, subject to the following conditions:
//
// The above copyright notice and this permission notice shall be included in
// all copies or substantial portions of the Software.
//
// THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS OR
// IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF MERCHANTABILITY,
// FITNESS FOR A PARTICULAR PURPOSE AND NONINFRINGEMENT.  IN NO EVENT SHALL THE
// AUTHORS OR COPYRIGHT HOLDERS BE LIABLE FOR ANY CLAIM, DAMAGES OR OTHER
// LIABILITY, WHETHER IN AN ACTION OF CONTRACT, TORT OR OTHERWISE, ARISING FROM,
// OUT OF OR IN CONNECTION WITH THE SOFTWARE OR THE USE OR OTHER DEALINGS IN
// THE SOFTWARE.

#include <gtest/gtest.h>
#include <stdio.h>

#include <cmath>
#include <type_traits>
#include <vector>

#include <hip/hip_runtime.h>

#include <rocrand/rocrand_kernel.h>

// GoogleTest-compatible HIP_CHECK macro. FAIL is called to log the Google Test trace.
// The lambda is invoked immediately as assertions that generate a fatal failure can
// only be used in void-returning functions.
#define HIP_CHECK(condition)                                                                \
    {                                                                                       \
        hipError_t error = condition;                                                       \
        if(error != hipSuccess)                                                             \
        {                                                                                   \
            [error]()                                                                       \
                { FAIL() << "HIP error " << error << ": " << hipGetErrorString(error); }(); \
            exit(error);                                                                    \
        }                                                                                   \
    }

#define ROCRAND_CHECK(state) ASSERT_EQ(state, ROCRAND_STATUS_SUCCESS)

template<class GeneratorState>
__global__
void rocrand_init_kernel(GeneratorState*               states,
                         const size_t                  states_size,
                         const unsigned long long int* vectors,
                         unsigned long long int        offset)
{
    const unsigned int state_id = blockIdx.x * blockDim.x + threadIdx.x;
    if(state_id < states_size)
    {
        GeneratorState state;
        rocrand_init(vectors, offset, &state);
        states[state_id] = state;
    }
}

template<class GeneratorState>
__global__
void rocrand_kernel(unsigned long long int*       output,
                    const unsigned long long int* vectors,
                    const size_t                  size)
{
    const unsigned int state_id    = blockIdx.x * blockDim.x + threadIdx.x;
    const unsigned int global_size = gridDim.x * blockDim.x;

    const unsigned int n = size / global_size;
    GeneratorState     state;
    rocrand_init(vectors, 1234 + state_id * n, &state);

    for(unsigned int i = 0; i < n; i++)
    {
        output[state_id * n + i] = rocrand(&state);
    }
}

template<class GeneratorState>
__global__
void rocrand_uniform_kernel(double*                       output,
                            const unsigned long long int* vectors,
                            const size_t                  size)
{
    const unsigned int state_id    = blockIdx.x * blockDim.x + threadIdx.x;
    const unsigned int global_size = gridDim.x * blockDim.x;

    const unsigned int n = size / global_size;
    GeneratorState     state;
    rocrand_init(vectors, 1234 + state_id * n, &state);

    for(unsigned int i = 0; i < n; i++)
    {
        output[state_id * n + i] = rocrand_uniform_double(&state);
    }
}

template<class GeneratorState>
__global__
void rocrand_normal_kernel(double* output, const unsigned long long int* vectors, const size_t size)
{
    const unsigned int state_id    = blockIdx.x * blockDim.x + threadIdx.x;
    const unsigned int global_size = gridDim.x * blockDim.x;

    const unsigned int n = size / global_size;
    GeneratorState     state;
    rocrand_init(vectors, 1234 + state_id * n, &state);

    for(unsigned int i = 0; i < n; i++)
    {
        output[state_id * n + i] = rocrand_normal_double(&state);
    }
}

template<class GeneratorState>
__global__
void rocrand_log_normal_kernel(double*                       output,
                               const unsigned long long int* vectors,
                               const size_t                  size)
{
    const unsigned int state_id    = blockIdx.x * blockDim.x + threadIdx.x;
    const unsigned int global_size = gridDim.x * blockDim.x;

    const unsigned int n = size / global_size;
    GeneratorState     state;
    rocrand_init(vectors, 1234 + state_id * n, &state);

    for(unsigned int i = 0; i < n; i++)
    {
        output[state_id * n + i] = rocrand_log_normal_double(&state, 1.6f, 0.25f);
    }
}

template<class GeneratorState>
__global__
void rocrand_poisson_kernel(unsigned int*                 output,
                            const unsigned long long int* vectors,
                            const size_t                  size,
                            double                        lambda)
{
    const unsigned int state_id    = blockIdx.x * blockDim.x + threadIdx.x;
    const unsigned int global_size = gridDim.x * blockDim.x;

    const unsigned int n = size / global_size;
    GeneratorState     state;
    rocrand_init(vectors, 1234 + state_id * n, &state);

    for(unsigned int i = 0; i < n; i++)
    {
        output[state_id * n + i] = rocrand_poisson(&state, lambda);
    }
}

template<class GeneratorState>
__global__
void rocrand_discrete_kernel(unsigned int*                 output,
                             const unsigned long long int* vectors,
                             const size_t                  size,
                             rocrand_discrete_distribution discrete_distribution)
{
    const unsigned int state_id    = blockIdx.x * blockDim.x + threadIdx.x;
    const unsigned int global_size = gridDim.x * blockDim.x;

    const unsigned int n = size / global_size;
    GeneratorState     state;
    rocrand_init(vectors, 1234 + state_id * n, &state);

    for(unsigned int i = 0; i < n; i++)
    {
        output[state_id * n + i] = rocrand_discrete(&state, discrete_distribution);
    }
}

TEST(rocrand_kernel_sobol64, rocrand_state_sobol64_type)
{
    typedef rocrand_state_sobol64 state_type;
    EXPECT_EQ(sizeof(state_type), 66 * sizeof(unsigned long long int));
    EXPECT_EQ(sizeof(state_type[32]), 32 * sizeof(state_type));
    EXPECT_TRUE(std::is_trivially_copyable<state_type>::value);
    EXPECT_TRUE(std::is_trivially_destructible<state_type>::value);
}

TEST(rocrand_kernel_sobol64, rocrand)
{
    typedef rocrand_state_sobol64 state_type;
    using Type = unsigned long long int;

    const size_t output_size = 8192;
    Type*        output;
    HIP_CHECK(hipMalloc(&output, output_size * sizeof(Type)));
    HIP_CHECK(hipDeviceSynchronize());

    const unsigned long long* h_directions;
    rocrand_get_direction_vectors64(&h_directions, ROCRAND_DIRECTION_VECTORS_64_JOEKUO6);

    Type* m_vector;
    HIP_CHECK(hipMalloc(&m_vector, sizeof(Type) * 8 * 64));
    HIP_CHECK(hipMemcpy(m_vector, h_directions, sizeof(Type) * 8 * 64, hipMemcpyHostToDevice));
    HIP_CHECK(hipDeviceSynchronize());

    hipLaunchKernelGGL(HIP_KERNEL_NAME(rocrand_kernel<state_type>),
                       dim3(8),
                       dim3(32),
                       0,
                       0,
                       output,
                       m_vector,
                       output_size);
    HIP_CHECK(hipGetLastError());

    std::vector<Type> output_host(output_size);
    HIP_CHECK(
        hipMemcpy(output_host.data(), output, output_size * sizeof(Type), hipMemcpyDeviceToHost));
    HIP_CHECK(hipDeviceSynchronize());
    HIP_CHECK(hipFree(output));
    HIP_CHECK(hipFree(m_vector));

    double mean = 0;
    for(auto v : output_host)
    {
        mean += static_cast<double>(v) / static_cast<double>(UINT64_MAX);
    }
    mean = mean / output_size;
    EXPECT_NEAR(mean, 0.5, 0.1);
}

TEST(rocrand_kernel_sobol64, rocrand_uniform)
{
    typedef rocrand_state_sobol64 state_type;
    typedef double                Type;

    const size_t output_size = 256;
    Type*        output;
    HIP_CHECK(hipMalloc(&output, output_size * sizeof(Type)));
    HIP_CHECK(hipDeviceSynchronize());

    typedef unsigned long long int DirectionVectorType;
    const DirectionVectorType*     h_directions;
    rocrand_get_direction_vectors64(&h_directions, ROCRAND_DIRECTION_VECTORS_64_JOEKUO6);

    DirectionVectorType* m_vector;
    HIP_CHECK(hipMalloc(&m_vector, sizeof(DirectionVectorType) * 8 * 64));
    HIP_CHECK(hipMemcpy(m_vector,
                        h_directions,
                        sizeof(DirectionVectorType) * 8 * 64,
                        hipMemcpyHostToDevice));
    HIP_CHECK(hipDeviceSynchronize());

    hipLaunchKernelGGL(HIP_KERNEL_NAME(rocrand_uniform_kernel<state_type>),
                       dim3(8),
                       dim3(32),
                       0,
                       0,
                       output,
                       m_vector,
                       output_size);
    HIP_CHECK(hipGetLastError());

    std::vector<Type> output_host(output_size);
    HIP_CHECK(
        hipMemcpy(output_host.data(), output, output_size * sizeof(Type), hipMemcpyDeviceToHost));
    HIP_CHECK(hipDeviceSynchronize());
    HIP_CHECK(hipFree(output));
    HIP_CHECK(hipFree(m_vector));

    double mean = 0;
    for(auto v : output_host)
    {
        mean += static_cast<double>(v);
    }
    mean = mean / output_size;
    EXPECT_NEAR(mean, 0.5, 0.1);
}

TEST(rocrand_kernel_sobol64, rocrand_normal)
{
    typedef rocrand_state_sobol64 state_type;
    typedef double                Type;

    const size_t output_size = 8192;
    Type*        output;
    HIP_CHECK(hipMalloc(&output, output_size * sizeof(Type)));
    HIP_CHECK(hipDeviceSynchronize());

    typedef unsigned long long int DirectionVectorType;
    const DirectionVectorType*     h_directions;
    rocrand_get_direction_vectors64(&h_directions, ROCRAND_DIRECTION_VECTORS_64_JOEKUO6);

    DirectionVectorType* m_vector;
    HIP_CHECK(hipMalloc(&m_vector, sizeof(DirectionVectorType) * 8 * 64));
    HIP_CHECK(hipMemcpy(m_vector,
                        h_directions,
                        sizeof(DirectionVectorType) * 8 * 64,
                        hipMemcpyHostToDevice));
    HIP_CHECK(hipDeviceSynchronize());

    hipLaunchKernelGGL(HIP_KERNEL_NAME(rocrand_normal_kernel<state_type>),
                       dim3(8),
                       dim3(32),
                       0,
                       0,
                       output,
                       m_vector,
                       output_size);
    HIP_CHECK(hipGetLastError());

    std::vector<Type> output_host(output_size);
    HIP_CHECK(
        hipMemcpy(output_host.data(), output, output_size * sizeof(Type), hipMemcpyDeviceToHost));
    HIP_CHECK(hipDeviceSynchronize());
    HIP_CHECK(hipFree(output));
    HIP_CHECK(hipFree(m_vector));

    Type mean = 0;
    for(auto v : output_host)
    {
        mean += static_cast<Type>(v);
    }
    mean = mean / output_size;
    EXPECT_NEAR(mean, 0.0, 0.2);

    Type stddev = 0;
    for(auto v : output_host)
    {
        stddev += std::pow(static_cast<Type>(v) - mean, 2);
    }
    stddev = stddev / output_size;
    EXPECT_NEAR(stddev, 1.0, 0.2);
}

TEST(rocrand_kernel_sobol64, rocrand_log_normal)
{
    typedef rocrand_state_sobol64 state_type;
    typedef double                Type;

    const size_t output_size = 8192;
    Type*        output;
    HIP_CHECK(hipMalloc(&output, output_size * sizeof(Type)));
    HIP_CHECK(hipDeviceSynchronize());

    typedef unsigned long long int DirectionVectorType;
    const DirectionVectorType*     h_directions;
    rocrand_get_direction_vectors64(&h_directions, ROCRAND_DIRECTION_VECTORS_64_JOEKUO6);

    DirectionVectorType* m_vector;
    HIP_CHECK(hipMalloc(&m_vector, sizeof(DirectionVectorType) * 8 * 64));
    HIP_CHECK(hipMemcpy(m_vector,
                        h_directions,
                        sizeof(DirectionVectorType) * 8 * 64,
                        hipMemcpyHostToDevice));
    HIP_CHECK(hipDeviceSynchronize());

    hipLaunchKernelGGL(HIP_KERNEL_NAME(rocrand_log_normal_kernel<state_type>),
                       dim3(8),
                       dim3(32),
                       0,
                       0,
                       output,
                       m_vector,
                       output_size);
    HIP_CHECK(hipGetLastError());

    std::vector<Type> output_host(output_size);
    HIP_CHECK(
        hipMemcpy(output_host.data(), output, output_size * sizeof(Type), hipMemcpyDeviceToHost));
    HIP_CHECK(hipDeviceSynchronize());
    HIP_CHECK(hipFree(output));
    HIP_CHECK(hipFree(m_vector));

    Type mean = 0;
    for(auto v : output_host)
    {
        mean += static_cast<Type>(v);
    }
    mean = mean / output_size;

    Type stddev = 0;
    for(auto v : output_host)
    {
        stddev += std::pow(v - mean, 2);
    }
    stddev = std::sqrt(stddev / output_size);

    Type logmean = std::log(mean * mean / std::sqrt(stddev + mean * mean));
    Type logstd  = std::sqrt(std::log(1.0f + stddev / (mean * mean)));

    EXPECT_NEAR(1.6, logmean, 1.6 * 0.2);
    EXPECT_NEAR(0.25, logstd, 0.25 * 0.2);
}

class rocrand_kernel_sobol64_poisson : public ::testing::TestWithParam<double>
{};

TEST_P(rocrand_kernel_sobol64_poisson, rocrand_poisson)
{
    typedef rocrand_state_sobol64 state_type;
    typedef double                Type;
    typedef unsigned int          ResultType;

    const Type lambda = GetParam();

    typedef unsigned long long int DirectionVectorType;
    const DirectionVectorType*     h_directions;
    rocrand_get_direction_vectors64(&h_directions, ROCRAND_DIRECTION_VECTORS_64_JOEKUO6);

    DirectionVectorType* m_vector;
    HIP_CHECK(hipMalloc(&m_vector, sizeof(DirectionVectorType) * 8 * 64));
    HIP_CHECK(hipMemcpy(m_vector,
                        h_directions,
                        sizeof(DirectionVectorType) * 8 * 64,
                        hipMemcpyHostToDevice));
    HIP_CHECK(hipDeviceSynchronize());

    const size_t output_size = 8192;
    ResultType*  output;
    HIP_CHECK(hipMalloc(&output, output_size * sizeof(ResultType)));
    HIP_CHECK(hipDeviceSynchronize());

    hipLaunchKernelGGL(HIP_KERNEL_NAME(rocrand_poisson_kernel<state_type>),
                       dim3(8),
                       dim3(32),
                       0,
                       0,
                       output,
                       m_vector,
                       output_size,
                       lambda);
    HIP_CHECK(hipGetLastError());

    std::vector<ResultType> output_host(output_size);
    HIP_CHECK(hipMemcpy(output_host.data(),
                        output,
                        output_size * sizeof(ResultType),
                        hipMemcpyDeviceToHost));
    HIP_CHECK(hipDeviceSynchronize());
    HIP_CHECK(hipFree(output));
    HIP_CHECK(hipFree(m_vector));

    Type mean = 0;
    for(auto v : output_host)
    {
        mean += static_cast<Type>(v);
    }
    mean = mean / output_size;

    Type variance = 0;
    for(auto v : output_host)
    {
        variance += std::pow(v - mean, 2);
    }
    variance = variance / output_size;

    EXPECT_NEAR(mean, lambda, std::max(1.0, lambda * 1e-1));
    EXPECT_NEAR(variance, lambda, std::max(1.0, lambda * 1e-1));
}

TEST_P(rocrand_kernel_sobol64_poisson, rocrand_discrete)
{
    typedef rocrand_state_sobol64 state_type;
    typedef double                Type;
    typedef unsigned int          ResultType;

    const Type lambda = GetParam();

    typedef unsigned long long int DirectionVectorType;
    const DirectionVectorType*     h_directions;
    rocrand_get_direction_vectors64(&h_directions, ROCRAND_DIRECTION_VECTORS_64_JOEKUO6);

    DirectionVectorType* m_vector;
    HIP_CHECK(hipMalloc(&m_vector, sizeof(DirectionVectorType) * 8 * 64));
    HIP_CHECK(hipMemcpy(m_vector,
                        h_directions,
                        sizeof(DirectionVectorType) * 8 * 64,
                        hipMemcpyHostToDevice));
    HIP_CHECK(hipDeviceSynchronize());

    const size_t output_size = 8192;
    ResultType*  output;
    HIP_CHECK(hipMalloc(&output, output_size * sizeof(ResultType)));
    HIP_CHECK(hipDeviceSynchronize());

    rocrand_discrete_distribution discrete_distribution;
    ROCRAND_CHECK(rocrand_create_poisson_distribution(lambda, &discrete_distribution));

    hipLaunchKernelGGL(HIP_KERNEL_NAME(rocrand_discrete_kernel<state_type>),
                       dim3(8),
                       dim3(32),
                       0,
                       0,
                       output,
                       m_vector,
                       output_size,
                       discrete_distribution);
    HIP_CHECK(hipGetLastError());

    std::vector<ResultType> output_host(output_size);
    HIP_CHECK(hipMemcpy(output_host.data(),
                        output,
                        output_size * sizeof(ResultType),
                        hipMemcpyDeviceToHost));
    HIP_CHECK(hipDeviceSynchronize());
    HIP_CHECK(hipFree(output));
    HIP_CHECK(hipFree(m_vector));
    ROCRAND_CHECK(rocrand_destroy_discrete_distribution(discrete_distribution));

    Type mean = 0;
    for(auto v : output_host)
    {
        mean += static_cast<Type>(v);
    }
    mean = mean / output_size;

    Type variance = 0;
    for(auto v : output_host)
    {
        variance += std::pow(v - mean, 2);
    }
    variance = variance / output_size;

    EXPECT_NEAR(mean, lambda, std::max(1.0, lambda * 1e-1));
    EXPECT_NEAR(variance, lambda, std::max(1.0, lambda * 1e-1));
}

const double lambdas[] = {1.0, 5.5, 20.0, 100.0, 1234.5, 5000.0};

INSTANTIATE_TEST_SUITE_P(rocrand_kernel_sobol64_poisson,
                         rocrand_kernel_sobol64_poisson,
                         ::testing::ValuesIn(lambdas));