Inverse.cpp

Go to the documentation of this file.

// ----------------------------------------------------------------------------
// -                        CloudViewer: www.cloudViewer.org                  -
// ----------------------------------------------------------------------------
// Copyright (c) 2018-2024 www.cloudViewer.org
// SPDX-License-Identifier: MIT
// ----------------------------------------------------------------------------
 
#include "cloudViewer/core/linalg/Inverse.h"
 
#include <unordered_map>
 
#include "cloudViewer/core/CUDAUtils.h"
#include "cloudViewer/core/linalg/LinalgHeadersCPU.h"
 
namespace cloudViewer {
namespace core {
 
void Inverse(const Tensor &A, Tensor &output) {
    AssertTensorDtypes(A, {Float32, Float64});
 
    const Device device = A.GetDevice();
    const Dtype dtype = A.GetDtype();
 
    // Check dimensions
    SizeVector A_shape = A.GetShape();
    if (A_shape.size() != 2) {
        utility::LogError("Tensor must be 2D, but got {}D.", A_shape.size());
    }
    if (A_shape[0] != A_shape[1]) {
        utility::LogError("Tensor must be square, but got {} x {}.", A_shape[0],
                          A_shape[1]);
    }
 
    int64_t n = A_shape[0];
    if (n == 0) {
        utility::LogError(
                "Tensor shapes should not contain dimensions with zero.");
    }
 
    if (device.IsCUDA()) {
#ifdef BUILD_CUDA_MODULE
        CUDAScopedDevice scoped_device(device);
        Tensor ipiv = Tensor::Zeros({n}, core::Int32, device);
        void *ipiv_data = ipiv.GetDataPtr();
 
        // cuSolver does not support getri, so we have to provide an identity
        // matrix. This matrix is modified in-place as output.
        Tensor A_T = A.T().Contiguous();
        void *A_data = A_T.GetDataPtr();
 
        output = Tensor::Eye(n, dtype, device);
        void *output_data = output.GetDataPtr();
 
        InverseCUDA(A_data, ipiv_data, output_data, n, dtype, device);
        output = output.T();
#else
        utility::LogError("Unimplemented device.");
#endif
    } else if (device.IsSYCL()) {
#ifdef BUILD_SYCL_MODULE
        Tensor ipiv = Tensor::Empty({n}, core::Int64, device);
        void *ipiv_data = ipiv.GetDataPtr();
 
        // LAPACKE supports getri, A is in-place modified as output.
        Tensor A_T = A.T().To(device, /*copy=*/true);
        void *A_data = A_T.GetDataPtr();
 
        InverseSYCL(A_data, ipiv_data, nullptr, n, dtype, device);
        output = A_T.T();
#else
        utility::LogError("Unimplemented device.");
#endif
    } else {
        Dtype ipiv_dtype;
        if (sizeof(CLOUDVIEWER_CPU_LINALG_INT) == 4) {
            ipiv_dtype = core::Int32;
        } else if (sizeof(CLOUDVIEWER_CPU_LINALG_INT) == 8) {
            ipiv_dtype = core::Int64;
        } else {
            utility::LogError("Unsupported CLOUDVIEWER_CPU_LINALG_INT type.");
        }
        Tensor ipiv = Tensor::Empty({n}, ipiv_dtype, device);
        void *ipiv_data = ipiv.GetDataPtr();
 
        // LAPACKE supports getri, A is in-place modified as output.
        Tensor A_T = A.T().To(device, /*copy=*/true);
        void *A_data = A_T.GetDataPtr();
 
        InverseCPU(A_data, ipiv_data, nullptr, n, dtype, device);
        output = A_T.T();
    }
}
}  // namespace core
}  // namespace cloudViewer