derek.holloway
7502bf64a4
CodeQL / Analyze (actions) (push) Canceled after 0s
CodeQL / Analyze (python) (push) Canceled after 0s
pre-commit / pre-commit (push) Canceled after 0s
Multi-Arch CI / setup (push) Canceled after 0s
Publish build_manylinux_x86_64 images / publish_build_manylinux_x86_64 (push) Canceled after 0s
Setup (Multi-Arch) / setup (push) Canceled after 0s
Multi-Arch CI / Linux::${{ fromJSON(needs.setup.outputs.linux_build_config || '{}').build_variant_label || 'skip' }} (push) Canceled after 0s
Publish TheRock Docker image / build-and-push-image (push) Canceled after 0s
Publish no_rocm_image_ubi10 image / publish_no_rocm_image_ubi10 (push) Canceled after 0s
Multi-Arch CI / Windows::${{ fromJSON(needs.setup.outputs.windows_build_config || '{}').build_variant_label || 'skip' }} (push) Canceled after 0s
Publish TheRock Docker image / build-and-push-image (push) Canceled after 0s
Multi-Arch CI / Manifest Diff (push) Canceled after 0s
TheRock Manifest Diff Report / Generate Manifest Diff Report (push) Canceled after 0s
Multi-Arch CI / CI Summary (push) Canceled after 0s
Publish no_rocm_image_ubuntu24_04 images / publish_no_rocm_image_ubuntu24_04 (push) Canceled after 0s
Publish no_rocm_image_ubuntu24_04_ocl_rt images / publish_no_rocm_image_ubuntu24_04_ocl_rt (push) Canceled after 0s
Publish TheRock Docker image / build-and-push-image (push) Canceled after 0s
Publish TheRock Docker image / build-and-push-image (push) Canceled after 0s
Publish no_rocm_image_ubuntu24_04 images / publish_no_rocm_image_ubuntu24_04_rocgdb (push) Canceled after 0s
Publish no_rocm_image_ubuntu24_04_rocgdb images / publish_no_rocm_image_ubuntu24_04_rocgdb (push) Canceled after 0s
Publish no_rocm_image_ubuntu24_04 images / publish_no_rocm_image_ubuntu24_04_ocl_rt (push) Canceled after 0s
Publish TheRock Docker image / build-and-push-image (push) Canceled after 0s
Unit Tests / Unit Tests :: ubuntu-24.04 (push) Canceled after 0s
Unit Tests / Unit Tests :: windows-2022 (push) Canceled after 0s
Publish no_rocm_image_ubuntu24_04 images / publish_no_rocm_image_ubuntu24_04_media (push) Canceled after 0s
Unit Tests / Unit Tests Summary (push) Canceled after 0s
Publish no_rocm_image_ubuntu24_04 images / publish_no_rocm_image_ubuntu24_04_openmpi (push) Canceled after 0s
1040 lines
47 KiB
C++
1040 lines
47 KiB
C++
/*! \file */
|
|
/* ************************************************************************
|
|
* Copyright (C) 2026 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 "testing_spildlt0.hpp"
|
|
#include "rocsparse_clients_objects.hpp"
|
|
#include "rocsparse_clients_spmat_descr.hpp"
|
|
#include "rocsparse_enum.hpp"
|
|
#include "testing.hpp"
|
|
|
|
// rocsparse_spildlt0_input_diag is deprecated but still exercised here (to check it keeps
|
|
// working). Referencing the [[deprecated]] enumerator directly would raise
|
|
// -Wdeprecated-declarations under -Wall, so use its integer value instead (mirrors
|
|
// deprecated_rocsparse_indextype_u16 in the library).
|
|
#define deprecated_rocsparse_spildlt0_input_diag static_cast<rocsparse_spildlt0_input>(7)
|
|
|
|
// Host reference for ILDLt factorization (templated on index types I, J)
|
|
template <typename T, typename I, typename J>
|
|
static void host_csrildlt0_ref(J M,
|
|
const I* __restrict__ csr_row_ptr,
|
|
const J* __restrict__ csr_col_ind,
|
|
T* __restrict__ csr_val,
|
|
floating_data_t<T>* __restrict__ diag,
|
|
rocsparse_index_base base,
|
|
int64_t* __restrict__ struct_pivot,
|
|
int64_t* __restrict__ numeric_pivot,
|
|
int64_t* __restrict__ singular_pivot,
|
|
double tol)
|
|
{
|
|
*struct_pivot = -1;
|
|
*numeric_pivot = -1;
|
|
*singular_pivot = -1;
|
|
|
|
std::vector<I> diag_offset(M, -1);
|
|
std::vector<I> nnz_entries(M, -1);
|
|
|
|
for(J ai = 0; ai < M; ++ai)
|
|
{
|
|
const I row_begin = csr_row_ptr[ai] - base;
|
|
const I row_end = csr_row_ptr[ai + 1] - base;
|
|
J j;
|
|
|
|
for(j = row_begin; j < row_end; ++j)
|
|
{
|
|
nnz_entries[csr_col_ind[j] - base] = j;
|
|
}
|
|
|
|
floating_data_t<T> diag_sum = static_cast<floating_data_t<T>>(0);
|
|
bool has_diag = false;
|
|
|
|
for(j = row_begin; j < row_end; ++j)
|
|
{
|
|
const J col_j = csr_col_ind[j] - base;
|
|
|
|
if(col_j == ai)
|
|
{
|
|
has_diag = true;
|
|
break;
|
|
}
|
|
|
|
if(col_j > ai)
|
|
{
|
|
break;
|
|
}
|
|
|
|
const floating_data_t<T> d_j = diag[col_j];
|
|
|
|
if(std::abs(d_j) <= tol)
|
|
{
|
|
*singular_pivot = (*singular_pivot == -1)
|
|
? (col_j + base)
|
|
: std::min(*singular_pivot, int64_t(col_j + base));
|
|
}
|
|
|
|
if(d_j == static_cast<floating_data_t<T>>(0))
|
|
{
|
|
*numeric_pivot = (*numeric_pivot == -1)
|
|
? (col_j + base)
|
|
: std::min(*numeric_pivot, int64_t(col_j + base));
|
|
continue;
|
|
}
|
|
|
|
const floating_data_t<T> inv_d_j = static_cast<floating_data_t<T>>(1) / d_j;
|
|
|
|
const I row_begin_j = csr_row_ptr[col_j] - base;
|
|
const I row_diag_j = diag_offset[col_j];
|
|
|
|
T local_sum = static_cast<T>(0);
|
|
|
|
for(I k = row_begin_j; k < row_diag_j; ++k)
|
|
{
|
|
const J col_k = csr_col_ind[k] - base;
|
|
if(nnz_entries[col_k] != -1)
|
|
{
|
|
const I idx = nnz_entries[col_k];
|
|
// L_{ai,k} * D_k * conj(L_{col_j,k})
|
|
local_sum = std::fma(csr_val[idx],
|
|
static_cast<T>(diag[col_k]) * rocsparse_conj(csr_val[k]),
|
|
local_sum);
|
|
}
|
|
}
|
|
|
|
const T val_j = (csr_val[j] - local_sum) * static_cast<T>(inv_d_j);
|
|
csr_val[j] = val_j;
|
|
|
|
// Accumulate |L_{ai,col_j}|^2 * D_{col_j} (real)
|
|
const floating_data_t<T> re_l = std::real(val_j);
|
|
const floating_data_t<T> im_l = std::imag(val_j);
|
|
diag_sum = std::fma(std::fma(re_l, re_l, im_l * im_l), d_j, diag_sum);
|
|
}
|
|
|
|
if(!has_diag)
|
|
{
|
|
*struct_pivot
|
|
= (*struct_pivot == -1) ? (ai + base) : std::min(*struct_pivot, int64_t(ai + base));
|
|
*numeric_pivot = (*numeric_pivot == -1) ? (ai + base)
|
|
: std::min(*numeric_pivot, int64_t(ai + base));
|
|
}
|
|
else
|
|
{
|
|
diag_offset[ai] = j;
|
|
|
|
const floating_data_t<T> d_i = std::real(csr_val[j]) - diag_sum;
|
|
|
|
if(std::abs(d_i) <= tol)
|
|
{
|
|
*singular_pivot = (*singular_pivot == -1)
|
|
? (ai + base)
|
|
: std::min(*singular_pivot, int64_t(ai + base));
|
|
}
|
|
|
|
if(d_i == static_cast<floating_data_t<T>>(0))
|
|
{
|
|
*numeric_pivot = (*numeric_pivot == -1)
|
|
? (ai + base)
|
|
: std::min(*numeric_pivot, int64_t(ai + base));
|
|
}
|
|
|
|
diag[ai] = d_i;
|
|
csr_val[j] = static_cast<T>(d_i); // D stored in-place on L's (implicit unit) diagonal
|
|
}
|
|
|
|
for(j = row_begin; j < row_end; ++j)
|
|
{
|
|
nnz_entries[csr_col_ind[j] - base] = -1;
|
|
}
|
|
}
|
|
|
|
if(*struct_pivot != -1)
|
|
{
|
|
*numeric_pivot
|
|
= (*numeric_pivot == -1) ? (*struct_pivot) : std::min(*numeric_pivot, *struct_pivot);
|
|
}
|
|
|
|
if(*numeric_pivot != -1)
|
|
{
|
|
*singular_pivot = (*singular_pivot == -1) ? (*numeric_pivot)
|
|
: std::min(*singular_pivot, *numeric_pivot);
|
|
}
|
|
}
|
|
|
|
namespace rocsparse_clients
|
|
{
|
|
class spildlt0_descr
|
|
{
|
|
private:
|
|
rocsparse_handle m_handle{};
|
|
rocsparse_spildlt0_descr m_descr{};
|
|
|
|
public:
|
|
host_dense_vector<int64_t> m_cpu_symbolic_singularity_position{};
|
|
host_dense_vector<int64_t> m_cpu_numeric_exact_singularity_position{};
|
|
host_dense_vector<int64_t> m_cpu_numeric_near_singularity_position{};
|
|
|
|
host_dense_vector<int64_t> m_host_symbolic_singularity_position{};
|
|
host_dense_vector<int64_t> m_host_numeric_singularity_position{};
|
|
|
|
device_dense_vector<int64_t> m_device_symbolic_singularity_position{};
|
|
device_dense_vector<int64_t> m_device_numeric_singularity_position{};
|
|
|
|
spildlt0_descr(rocsparse_handle handle, int64_t batch_count)
|
|
: m_handle(handle)
|
|
, m_cpu_symbolic_singularity_position(batch_count)
|
|
, m_cpu_numeric_exact_singularity_position(batch_count)
|
|
, m_cpu_numeric_near_singularity_position(batch_count)
|
|
, m_host_symbolic_singularity_position(batch_count)
|
|
, m_host_numeric_singularity_position(batch_count)
|
|
, m_device_symbolic_singularity_position(batch_count)
|
|
, m_device_numeric_singularity_position(batch_count)
|
|
{
|
|
ROCSPARSE_CLIENTS_ROUTINE_TRACE;
|
|
rocsparse_error* p_error = nullptr;
|
|
const rocsparse_status status
|
|
= rocsparse_spildlt0_descr_create(handle, &this->m_descr, p_error);
|
|
if(status != rocsparse_status_success)
|
|
{
|
|
throw(status);
|
|
}
|
|
}
|
|
|
|
~spildlt0_descr()
|
|
{
|
|
ROCSPARSE_CLIENTS_ROUTINE_TRACE;
|
|
rocsparse_error* p_error = nullptr;
|
|
std::ignore = rocsparse_spildlt0_descr_destroy(this->m_handle, this->m_descr, p_error);
|
|
}
|
|
|
|
inline operator rocsparse_spildlt0_descr&()
|
|
{
|
|
return this->m_descr;
|
|
}
|
|
|
|
inline operator const rocsparse_spildlt0_descr&() const
|
|
{
|
|
return this->m_descr;
|
|
}
|
|
};
|
|
}
|
|
|
|
template <typename I, typename J, typename T>
|
|
void testing_spildlt0_set_input_bad_arg(const Arguments& arg)
|
|
{
|
|
rocsparse_error* p_error = nullptr;
|
|
rocsparse_local_handle local_handle;
|
|
rocsparse_spildlt0_descr spildlt0_descr = (rocsparse_spildlt0_descr)0x4;
|
|
rocsparse_handle handle = local_handle;
|
|
static constexpr int nex = 2;
|
|
static const int ex[nex] = {4, 5};
|
|
const rocsparse_spildlt0_input input = rocsparse_spildlt0_input_alg;
|
|
void* data = (void*)0x4;
|
|
size_t data_size_in_bytes = sizeof(input);
|
|
select_bad_arg_analysis(rocsparse_spildlt0_set_input,
|
|
nex,
|
|
ex,
|
|
handle, //0
|
|
spildlt0_descr, //1
|
|
input, //2
|
|
data, //3
|
|
data_size_in_bytes, //4
|
|
p_error); // 5
|
|
|
|
CHECK_ROCSPARSE_ERROR(rocsparse_spildlt0_descr_create(handle, &spildlt0_descr, p_error));
|
|
for(auto e : {rocsparse_spildlt0_input_alg,
|
|
rocsparse_spildlt0_input_analysis_policy,
|
|
rocsparse_spildlt0_input_compute_datatype,
|
|
rocsparse_spildlt0_input_boost_enable,
|
|
rocsparse_spildlt0_input_boost_tolerance,
|
|
rocsparse_spildlt0_input_boost_value,
|
|
rocsparse_spildlt0_input_singularity_tolerance,
|
|
deprecated_rocsparse_spildlt0_input_diag})
|
|
{
|
|
switch(e)
|
|
{
|
|
case rocsparse_spildlt0_input_alg:
|
|
case rocsparse_spildlt0_input_analysis_policy:
|
|
case rocsparse_spildlt0_input_compute_datatype:
|
|
case rocsparse_spildlt0_input_boost_enable:
|
|
case rocsparse_spildlt0_input_boost_tolerance:
|
|
case rocsparse_spildlt0_input_boost_value:
|
|
case rocsparse_spildlt0_input_singularity_tolerance:
|
|
case deprecated_rocsparse_spildlt0_input_diag:
|
|
{
|
|
EXPECT_ROCSPARSE_STATUS(
|
|
rocsparse_spildlt0_set_input(handle, spildlt0_descr, e, data, 0, p_error),
|
|
rocsparse_status_invalid_value);
|
|
break;
|
|
}
|
|
}
|
|
}
|
|
|
|
//
|
|
// Singularity tolerance must be double
|
|
//
|
|
EXPECT_ROCSPARSE_STATUS(
|
|
rocsparse_spildlt0_set_input(handle,
|
|
spildlt0_descr,
|
|
rocsparse_spildlt0_input_singularity_tolerance,
|
|
data,
|
|
sizeof(float),
|
|
p_error),
|
|
rocsparse_status_invalid_value);
|
|
|
|
CHECK_ROCSPARSE_ERROR(rocsparse_spildlt0_descr_destroy(handle, spildlt0_descr, p_error));
|
|
}
|
|
|
|
template <typename I, typename J, typename T>
|
|
void testing_spildlt0_buffer_size_bad_arg(const Arguments& arg)
|
|
{
|
|
rocsparse_local_handle local_handle;
|
|
|
|
rocsparse_handle handle = local_handle;
|
|
rocsparse_spildlt0_descr spildlt0_descr = (rocsparse_spildlt0_descr)0x4;
|
|
|
|
rocsparse_spmat_descr A = (rocsparse_spmat_descr)0x4;
|
|
rocsparse_spmat_descr P = (rocsparse_spmat_descr)0x4;
|
|
rocsparse_spildlt0_stage spildlt0_stage = rocsparse_spildlt0_stage_analysis;
|
|
|
|
rocsparse_error* p_error = nullptr;
|
|
{
|
|
size_t* p_buffer_size_in_bytes = (size_t*)0x4;
|
|
static constexpr int nex = 1;
|
|
static const int ex[nex] = {6};
|
|
select_bad_arg_analysis(rocsparse_spildlt0_buffer_size,
|
|
nex,
|
|
ex,
|
|
handle,
|
|
spildlt0_descr,
|
|
A,
|
|
P,
|
|
spildlt0_stage,
|
|
p_buffer_size_in_bytes,
|
|
p_error);
|
|
}
|
|
}
|
|
|
|
template <typename I, typename J, typename T>
|
|
void testing_spildlt0_get_output_bad_arg(const Arguments& arg)
|
|
{
|
|
rocsparse_local_handle local_handle;
|
|
rocsparse_handle handle = local_handle;
|
|
rocsparse_spildlt0_descr spildlt0_descr = (rocsparse_spildlt0_descr)0x4;
|
|
rocsparse_error* p_error = nullptr;
|
|
static constexpr int nex = 2;
|
|
static const int ex[nex] = {4, 5};
|
|
size_t data_size_in_bytes = sizeof(int64_t);
|
|
void* data = (void*)0x4;
|
|
rocsparse_spildlt0_output output = rocsparse_spildlt0_output_singularity_position;
|
|
select_bad_arg_analysis(rocsparse_spildlt0_get_output,
|
|
nex,
|
|
ex,
|
|
handle,
|
|
spildlt0_descr,
|
|
output,
|
|
data,
|
|
data_size_in_bytes,
|
|
p_error);
|
|
}
|
|
|
|
template <typename I, typename J, typename T>
|
|
void testing_spildlt0_analysis_bad_arg(const Arguments& arg)
|
|
{
|
|
rocsparse_local_handle local_handle;
|
|
rocsparse_handle handle = local_handle;
|
|
rocsparse_spildlt0_descr spildlt0_descr = (rocsparse_spildlt0_descr)0x4;
|
|
rocsparse_spmat_descr A = (rocsparse_spmat_descr)0x4;
|
|
rocsparse_spmat_descr P = (rocsparse_spmat_descr)0x4;
|
|
rocsparse_spildlt0_stage spildlt0_stage = rocsparse_spildlt0_stage_analysis;
|
|
void* buffer = (void*)0x4;
|
|
rocsparse_error* p_error = nullptr;
|
|
{
|
|
const size_t buffer_size_in_bytes = 1;
|
|
#define PARAMS handle, spildlt0_descr, A, P, spildlt0_stage, buffer_size_in_bytes, buffer, p_error
|
|
static constexpr int nex = 2;
|
|
static const int ex[nex] = {5, 7};
|
|
select_bad_arg_analysis(rocsparse_spildlt0, nex, ex, PARAMS);
|
|
#undef PARAMS
|
|
}
|
|
}
|
|
|
|
template <typename I, typename J, typename T>
|
|
void testing_spildlt0_bad_arg(const Arguments& arg)
|
|
{
|
|
testing_spildlt0_buffer_size_bad_arg<I, J, T>(arg);
|
|
testing_spildlt0_analysis_bad_arg<I, J, T>(arg);
|
|
testing_spildlt0_set_input_bad_arg<I, J, T>(arg);
|
|
testing_spildlt0_get_output_bad_arg<I, J, T>(arg);
|
|
|
|
rocsparse_error* p_error = nullptr;
|
|
rocsparse_local_handle local_handle;
|
|
rocsparse_handle handle = local_handle;
|
|
|
|
//
|
|
// Now get a concrete example and continue.
|
|
//
|
|
rocsparse_clients::csr_tridiag_matrix_t<T, I, J> A(4);
|
|
rocsparse_spmat_descr P = A;
|
|
host_scalar<double> hsingularity_tolerance(1);
|
|
hsingularity_tolerance[0] = 1.0e-5;
|
|
device_scalar<double> dsingularity_tolerance(hsingularity_tolerance);
|
|
|
|
// Allocate a diag vector on device (size = 4, same as tridiag matrix dimension)
|
|
device_dense_vector<floating_data_t<T>> d_diag(4);
|
|
|
|
for(auto analysis_policy : {rocsparse_analysis_policy_force, rocsparse_analysis_policy_reuse})
|
|
{
|
|
rocsparse_clients::spildlt0_descr spildlt0_descr(handle, 1);
|
|
|
|
const rocsparse_spildlt0_alg alg = rocsparse_spildlt0_alg_default;
|
|
const rocsparse_datatype compute_datatype = get_datatype<T>();
|
|
CHECK_ROCSPARSE_ERROR(rocsparse_spildlt0_set_input(
|
|
handle, spildlt0_descr, rocsparse_spildlt0_input_alg, &alg, sizeof(alg), p_error));
|
|
|
|
CHECK_ROCSPARSE_ERROR(rocsparse_spildlt0_set_input(handle,
|
|
spildlt0_descr,
|
|
rocsparse_spildlt0_input_analysis_policy,
|
|
&analysis_policy,
|
|
sizeof(analysis_policy),
|
|
p_error));
|
|
|
|
CHECK_ROCSPARSE_ERROR(
|
|
rocsparse_spildlt0_set_input(handle,
|
|
spildlt0_descr,
|
|
rocsparse_spildlt0_input_compute_datatype,
|
|
&compute_datatype,
|
|
sizeof(compute_datatype),
|
|
p_error));
|
|
|
|
// Set the diag device pointer
|
|
void* d_diag_ptr = (void*)d_diag;
|
|
CHECK_ROCSPARSE_ERROR(rocsparse_spildlt0_set_input(handle,
|
|
spildlt0_descr,
|
|
deprecated_rocsparse_spildlt0_input_diag,
|
|
&d_diag_ptr,
|
|
sizeof(d_diag_ptr),
|
|
p_error));
|
|
|
|
size_t buffer_size;
|
|
CHECK_ROCSPARSE_ERROR(rocsparse_spildlt0_buffer_size(handle,
|
|
spildlt0_descr,
|
|
A,
|
|
P,
|
|
rocsparse_spildlt0_stage_analysis,
|
|
&buffer_size,
|
|
p_error));
|
|
hipStream_t stream;
|
|
CHECK_ROCSPARSE_ERROR(rocsparse_get_stream(handle, &stream));
|
|
{
|
|
device_dense_vector<char> buffer(buffer_size);
|
|
CHECK_HIP_ERROR(hipMemset(buffer, 255 - 1, buffer_size));
|
|
|
|
//
|
|
// Call compute before analysis.
|
|
//
|
|
EXPECT_ROCSPARSE_STATUS(rocsparse_status_invalid_value,
|
|
rocsparse_spildlt0(handle,
|
|
spildlt0_descr,
|
|
A,
|
|
P,
|
|
rocsparse_spildlt0_stage_compute,
|
|
buffer_size,
|
|
buffer,
|
|
p_error));
|
|
|
|
//
|
|
// Call analysis.
|
|
//
|
|
CHECK_ROCSPARSE_ERROR(rocsparse_spildlt0(handle,
|
|
spildlt0_descr,
|
|
A,
|
|
P,
|
|
rocsparse_spildlt0_stage_analysis,
|
|
buffer_size,
|
|
buffer,
|
|
p_error));
|
|
|
|
for(auto input : {rocsparse_spildlt0_input_alg,
|
|
rocsparse_spildlt0_input_analysis_policy,
|
|
rocsparse_spildlt0_input_compute_datatype,
|
|
rocsparse_spildlt0_input_boost_enable,
|
|
rocsparse_spildlt0_input_boost_tolerance,
|
|
rocsparse_spildlt0_input_boost_value,
|
|
rocsparse_spildlt0_input_singularity_tolerance,
|
|
deprecated_rocsparse_spildlt0_input_diag})
|
|
{
|
|
switch(input)
|
|
{
|
|
case rocsparse_spildlt0_input_alg:
|
|
case rocsparse_spildlt0_input_compute_datatype:
|
|
case rocsparse_spildlt0_input_analysis_policy:
|
|
{
|
|
EXPECT_ROCSPARSE_STATUS(
|
|
rocsparse_status_invalid_value,
|
|
rocsparse_spildlt0_set_input(
|
|
handle, spildlt0_descr, input, (void*)0x4, sizeof(int64_t), p_error));
|
|
break;
|
|
}
|
|
case rocsparse_spildlt0_input_singularity_tolerance:
|
|
{
|
|
EXPECT_ROCSPARSE_STATUS(
|
|
rocsparse_status_invalid_value,
|
|
rocsparse_spildlt0_set_input(
|
|
handle, spildlt0_descr, input, (void*)0x4, sizeof(float), p_error));
|
|
|
|
CHECK_ROCSPARSE_ERROR(
|
|
rocsparse_set_pointer_mode(handle, rocsparse_pointer_mode_host));
|
|
CHECK_ROCSPARSE_ERROR(
|
|
rocsparse_spildlt0_set_input(handle,
|
|
spildlt0_descr,
|
|
rocsparse_spildlt0_input_singularity_tolerance,
|
|
hsingularity_tolerance,
|
|
sizeof(double),
|
|
p_error));
|
|
CHECK_ROCSPARSE_ERROR(
|
|
rocsparse_set_pointer_mode(handle, rocsparse_pointer_mode_device));
|
|
CHECK_ROCSPARSE_ERROR(
|
|
rocsparse_spildlt0_set_input(handle,
|
|
spildlt0_descr,
|
|
rocsparse_spildlt0_input_singularity_tolerance,
|
|
dsingularity_tolerance,
|
|
sizeof(double),
|
|
p_error));
|
|
break;
|
|
}
|
|
case rocsparse_spildlt0_input_boost_enable:
|
|
case rocsparse_spildlt0_input_boost_tolerance:
|
|
case rocsparse_spildlt0_input_boost_value:
|
|
case deprecated_rocsparse_spildlt0_input_diag:
|
|
{
|
|
break;
|
|
}
|
|
}
|
|
}
|
|
|
|
//
|
|
// Call analysis twice.
|
|
//
|
|
EXPECT_ROCSPARSE_STATUS(rocsparse_status_invalid_value,
|
|
rocsparse_spildlt0(handle,
|
|
spildlt0_descr,
|
|
A,
|
|
P,
|
|
rocsparse_spildlt0_stage_analysis,
|
|
buffer_size,
|
|
buffer,
|
|
p_error));
|
|
}
|
|
|
|
{
|
|
CHECK_ROCSPARSE_ERROR(rocsparse_spildlt0_buffer_size(handle,
|
|
spildlt0_descr,
|
|
A,
|
|
P,
|
|
rocsparse_spildlt0_stage_compute,
|
|
&buffer_size,
|
|
p_error));
|
|
device_dense_vector<char> buffer(buffer_size);
|
|
CHECK_HIP_ERROR(hipMemset(buffer, 255 - 1, buffer_size));
|
|
|
|
CHECK_ROCSPARSE_ERROR(rocsparse_set_pointer_mode(handle, rocsparse_pointer_mode_host));
|
|
CHECK_ROCSPARSE_ERROR(
|
|
rocsparse_spildlt0_set_input(handle,
|
|
spildlt0_descr,
|
|
rocsparse_spildlt0_input_singularity_tolerance,
|
|
hsingularity_tolerance,
|
|
sizeof(double),
|
|
p_error));
|
|
|
|
//
|
|
// Call compute.
|
|
//
|
|
CHECK_ROCSPARSE_ERROR(rocsparse_spildlt0(handle,
|
|
spildlt0_descr,
|
|
A,
|
|
P,
|
|
rocsparse_spildlt0_stage_compute,
|
|
buffer_size,
|
|
buffer,
|
|
p_error));
|
|
|
|
//
|
|
// Call analysis after compute.
|
|
//
|
|
EXPECT_ROCSPARSE_STATUS(rocsparse_status_invalid_value,
|
|
rocsparse_spildlt0(handle,
|
|
spildlt0_descr,
|
|
A,
|
|
P,
|
|
rocsparse_spildlt0_stage_analysis,
|
|
buffer_size,
|
|
buffer,
|
|
p_error));
|
|
|
|
CHECK_ROCSPARSE_ERROR(
|
|
rocsparse_set_pointer_mode(handle, rocsparse_pointer_mode_device));
|
|
CHECK_ROCSPARSE_ERROR(
|
|
rocsparse_spildlt0_set_input(handle,
|
|
spildlt0_descr,
|
|
rocsparse_spildlt0_input_singularity_tolerance,
|
|
dsingularity_tolerance,
|
|
sizeof(double),
|
|
p_error));
|
|
|
|
//
|
|
// Call compute.
|
|
//
|
|
CHECK_ROCSPARSE_ERROR(rocsparse_spildlt0(handle,
|
|
spildlt0_descr,
|
|
A,
|
|
P,
|
|
rocsparse_spildlt0_stage_compute,
|
|
buffer_size,
|
|
buffer,
|
|
p_error));
|
|
|
|
//
|
|
// Call analysis after compute.
|
|
//
|
|
EXPECT_ROCSPARSE_STATUS(rocsparse_status_invalid_value,
|
|
rocsparse_spildlt0(handle,
|
|
spildlt0_descr,
|
|
A,
|
|
P,
|
|
rocsparse_spildlt0_stage_analysis,
|
|
buffer_size,
|
|
buffer,
|
|
p_error));
|
|
//
|
|
// Call compute.
|
|
//
|
|
CHECK_ROCSPARSE_ERROR(rocsparse_spildlt0(handle,
|
|
spildlt0_descr,
|
|
A,
|
|
P,
|
|
rocsparse_spildlt0_stage_compute,
|
|
buffer_size,
|
|
buffer,
|
|
p_error));
|
|
|
|
CHECK_HIP_ERROR(hipStreamSynchronize(stream));
|
|
}
|
|
}
|
|
}
|
|
|
|
template <typename I, typename J, typename T>
|
|
void testing_spildlt0(const Arguments& arg)
|
|
{
|
|
rocsparse_error* p_error = nullptr;
|
|
|
|
if(arg.M != arg.N)
|
|
{
|
|
return;
|
|
}
|
|
|
|
int64_t batch_count = arg.batch_count;
|
|
if(batch_count == -1)
|
|
batch_count = 1;
|
|
|
|
static constexpr const bool full_rank = true;
|
|
rocsparse_clients::spmat_descr<T, I, J> A(arg, batch_count, full_rank);
|
|
|
|
if(false == A.is_square())
|
|
{
|
|
return;
|
|
}
|
|
|
|
// Only CSR is supported for spildlt0
|
|
if(A.get_format() != rocsparse_format_csr)
|
|
{
|
|
return;
|
|
}
|
|
|
|
rocsparse_local_handle handle(arg);
|
|
hipStream_t stream;
|
|
CHECK_ROCSPARSE_ERROR(rocsparse_get_stream(handle, &stream));
|
|
|
|
const int64_t M = A.get_nrows();
|
|
|
|
rocsparse_clients::spildlt0_descr spildlt0_descr(handle, batch_count);
|
|
|
|
const rocsparse_spildlt0_alg alg = rocsparse_spildlt0_alg_default;
|
|
const rocsparse_analysis_policy analysis_policy = arg.apol;
|
|
const rocsparse_datatype compute_datatype = get_datatype<T>();
|
|
|
|
CHECK_ROCSPARSE_ERROR(rocsparse_spildlt0_set_input(
|
|
handle, spildlt0_descr, rocsparse_spildlt0_input_alg, &alg, sizeof(alg), p_error));
|
|
|
|
CHECK_ROCSPARSE_ERROR(rocsparse_spildlt0_set_input(handle,
|
|
spildlt0_descr,
|
|
rocsparse_spildlt0_input_analysis_policy,
|
|
&analysis_policy,
|
|
sizeof(analysis_policy),
|
|
p_error));
|
|
|
|
CHECK_ROCSPARSE_ERROR(rocsparse_spildlt0_set_input(handle,
|
|
spildlt0_descr,
|
|
rocsparse_spildlt0_input_compute_datatype,
|
|
&compute_datatype,
|
|
sizeof(compute_datatype),
|
|
p_error));
|
|
|
|
//
|
|
// Perform analysis.
|
|
//
|
|
auto& host_symbolic_pivot = spildlt0_descr.m_host_symbolic_singularity_position;
|
|
|
|
{
|
|
size_t buffer_size_in_bytes = std::numeric_limits<size_t>::max();
|
|
CHECK_ROCSPARSE_ERROR(rocsparse_spildlt0_buffer_size(handle,
|
|
spildlt0_descr,
|
|
A,
|
|
A,
|
|
rocsparse_spildlt0_stage_analysis,
|
|
&buffer_size_in_bytes,
|
|
p_error));
|
|
device_dense_vector<char> buffer(buffer_size_in_bytes);
|
|
CHECK_HIP_ERROR(hipMemset(buffer, 255 - 1, buffer_size_in_bytes));
|
|
|
|
CHECK_ROCSPARSE_ERROR(rocsparse_spildlt0(handle,
|
|
spildlt0_descr,
|
|
A,
|
|
A,
|
|
rocsparse_spildlt0_stage_analysis,
|
|
buffer_size_in_bytes,
|
|
buffer,
|
|
p_error));
|
|
|
|
CHECK_ROCSPARSE_ERROR(rocsparse_set_pointer_mode(handle, rocsparse_pointer_mode_host));
|
|
|
|
CHECK_ROCSPARSE_ERROR(
|
|
rocsparse_spildlt0_get_output(handle,
|
|
spildlt0_descr,
|
|
rocsparse_spildlt0_output_singularity_position,
|
|
host_symbolic_pivot,
|
|
sizeof(int64_t),
|
|
p_error));
|
|
|
|
CHECK_HIP_ERROR(hipStreamSynchronize(stream));
|
|
}
|
|
|
|
auto& device_symbolic_pivot = spildlt0_descr.m_device_symbolic_singularity_position;
|
|
auto& host_numeric_pivot = spildlt0_descr.m_host_numeric_singularity_position;
|
|
auto& device_numeric_pivot = spildlt0_descr.m_device_numeric_singularity_position;
|
|
|
|
if(arg.unit_check)
|
|
{
|
|
CHECK_ROCSPARSE_ERROR(rocsparse_set_pointer_mode(handle, rocsparse_pointer_mode_device));
|
|
|
|
CHECK_ROCSPARSE_ERROR(
|
|
rocsparse_spildlt0_get_output(handle,
|
|
spildlt0_descr,
|
|
rocsparse_spildlt0_output_singularity_position,
|
|
device_symbolic_pivot,
|
|
sizeof(int64_t),
|
|
p_error));
|
|
|
|
CHECK_HIP_ERROR(hipStreamSynchronize(stream));
|
|
|
|
// Check pivot is the same with either host or device mode.
|
|
host_symbolic_pivot.unit_check(device_symbolic_pivot);
|
|
|
|
//
|
|
// Perform compute.
|
|
//
|
|
{
|
|
size_t buffer_size_in_bytes = std::numeric_limits<size_t>::max();
|
|
CHECK_ROCSPARSE_ERROR(rocsparse_spildlt0_buffer_size(handle,
|
|
spildlt0_descr,
|
|
A,
|
|
A,
|
|
rocsparse_spildlt0_stage_compute,
|
|
&buffer_size_in_bytes,
|
|
p_error));
|
|
|
|
device_dense_vector<char> buffer(buffer_size_in_bytes);
|
|
CHECK_HIP_ERROR(hipMemset(buffer, 255 - 1, buffer_size_in_bytes));
|
|
|
|
CHECK_ROCSPARSE_ERROR(rocsparse_spildlt0(handle,
|
|
spildlt0_descr,
|
|
A,
|
|
A,
|
|
rocsparse_spildlt0_stage_compute,
|
|
buffer_size_in_bytes,
|
|
buffer,
|
|
p_error));
|
|
|
|
CHECK_ROCSPARSE_ERROR(rocsparse_set_pointer_mode(handle, rocsparse_pointer_mode_host));
|
|
|
|
CHECK_ROCSPARSE_ERROR(
|
|
rocsparse_spildlt0_get_output(handle,
|
|
spildlt0_descr,
|
|
rocsparse_spildlt0_output_singularity_position,
|
|
host_numeric_pivot,
|
|
sizeof(int64_t),
|
|
p_error));
|
|
|
|
CHECK_HIP_ERROR(hipStreamSynchronize(stream));
|
|
|
|
CHECK_ROCSPARSE_ERROR(
|
|
rocsparse_set_pointer_mode(handle, rocsparse_pointer_mode_device));
|
|
|
|
CHECK_ROCSPARSE_ERROR(
|
|
rocsparse_spildlt0_get_output(handle,
|
|
spildlt0_descr,
|
|
rocsparse_spildlt0_output_singularity_position,
|
|
device_numeric_pivot,
|
|
sizeof(int64_t),
|
|
p_error));
|
|
|
|
CHECK_HIP_ERROR(hipStreamSynchronize(stream));
|
|
|
|
// Check numeric pivots match.
|
|
host_numeric_pivot.unit_check(device_numeric_pivot);
|
|
|
|
// Perform calculation on host and compare.
|
|
{
|
|
auto& host_csr = A.template as<rocsparse_format_csr>().host();
|
|
|
|
// Copy the factored device values back to host for comparison. D is stored
|
|
// in-place on L's diagonal, so this single array carries both L and D: the
|
|
// value comparison below validates D on the diagonal without any host-side
|
|
// gather.
|
|
host_vector<T> h_csr_val_gpu(host_csr.val.size());
|
|
CHECK_HIP_ERROR(hipMemcpy(h_csr_val_gpu,
|
|
A.get_device_values(),
|
|
sizeof(T) * host_csr.val.size(),
|
|
hipMemcpyDeviceToHost));
|
|
|
|
// CPU reference (run on a copy of the original host values)
|
|
host_vector<T> h_csr_val_ref = host_csr.val;
|
|
host_vector<floating_data_t<T>> h_diag_ref(M * batch_count,
|
|
static_cast<floating_data_t<T>>(0));
|
|
|
|
const double singular_pivot_tolerance = 0.0;
|
|
|
|
auto& cpu_symbolic_pivot = spildlt0_descr.m_cpu_symbolic_singularity_position;
|
|
auto& cpu_numeric_near_pivot
|
|
= spildlt0_descr.m_cpu_numeric_near_singularity_position;
|
|
auto& cpu_numeric_exact_pivot
|
|
= spildlt0_descr.m_cpu_numeric_exact_singularity_position;
|
|
|
|
for(int64_t i = 0; i < batch_count; ++i)
|
|
{
|
|
host_csrildlt0_ref<T, I, J>(static_cast<J>(host_csr.m),
|
|
host_csr.ptr.data(),
|
|
host_csr.ind.data(),
|
|
h_csr_val_ref.data() + i * A.get_stride(),
|
|
h_diag_ref.data() + i * M,
|
|
host_csr.base,
|
|
cpu_symbolic_pivot + i,
|
|
cpu_numeric_exact_pivot + i,
|
|
cpu_numeric_near_pivot + i,
|
|
singular_pivot_tolerance);
|
|
}
|
|
|
|
for(int64_t j = 0; j < batch_count; ++j)
|
|
{
|
|
if(cpu_numeric_near_pivot[j] == -1)
|
|
{
|
|
cpu_numeric_near_pivot[j] = cpu_numeric_exact_pivot[j];
|
|
}
|
|
else if(cpu_numeric_exact_pivot[j] != -1)
|
|
{
|
|
cpu_numeric_near_pivot[j]
|
|
= std::min(cpu_numeric_near_pivot[j], cpu_numeric_exact_pivot[j]);
|
|
}
|
|
}
|
|
|
|
cpu_symbolic_pivot.unit_check(host_symbolic_pivot);
|
|
cpu_numeric_near_pivot.unit_check(host_numeric_pivot);
|
|
|
|
// Check values only when no structural/numeric zero pivot was encountered (all batches).
|
|
bool no_pivot = true;
|
|
for(int64_t i = 0; i < batch_count; ++i)
|
|
{
|
|
if(host_symbolic_pivot[i] != -1 || host_numeric_pivot[i] != -1)
|
|
{
|
|
no_pivot = false;
|
|
break;
|
|
}
|
|
}
|
|
if(no_pivot)
|
|
{
|
|
// For complex types, accumulation error is larger.
|
|
constexpr bool is_complex = std::is_same<T, rocsparse_float_complex>::value
|
|
|| std::is_same<T, rocsparse_double_complex>::value;
|
|
const floating_data_t<T> tol_ildlt
|
|
= is_complex
|
|
? static_cast<floating_data_t<T>>(100) * default_tolerance<T>::value
|
|
: default_tolerance<T>::value;
|
|
|
|
// Compare L values (off-diagonal CSR entries, plus D on the diagonal). This
|
|
// validates D-on-diagonal directly, with nothing gathered on the host.
|
|
h_csr_val_ref.near_check(h_csr_val_gpu, tol_ildlt);
|
|
|
|
// Validate the OPTIONAL external diagonal output vector, which is a *device*
|
|
// buffer filled by the kernel (no host round-trip to produce D). Re-run the
|
|
// compute with it set to also exercise the per-batch copy: each batch must
|
|
// receive its own D into its own m-entry slice (offset b * m).
|
|
device_dense_vector<floating_data_t<T>> d_diag(M * batch_count);
|
|
{
|
|
// Compute is in-place and destructive: restore the original values.
|
|
// host_csr.val is still pristine (the reference ran on h_csr_val_ref).
|
|
A.get_device_values().transfer_from(host_csr.val);
|
|
|
|
void* d_diag_ptr = (void*)d_diag;
|
|
CHECK_ROCSPARSE_ERROR(
|
|
rocsparse_spildlt0_set_input(handle,
|
|
spildlt0_descr,
|
|
deprecated_rocsparse_spildlt0_input_diag,
|
|
&d_diag_ptr,
|
|
sizeof(d_diag_ptr),
|
|
p_error));
|
|
|
|
CHECK_ROCSPARSE_ERROR(rocsparse_spildlt0(handle,
|
|
spildlt0_descr,
|
|
A,
|
|
A,
|
|
rocsparse_spildlt0_stage_compute,
|
|
buffer_size_in_bytes,
|
|
buffer,
|
|
p_error));
|
|
CHECK_HIP_ERROR(hipStreamSynchronize(stream));
|
|
}
|
|
|
|
// Copy the device D vector to host only for the reference comparison.
|
|
host_vector<floating_data_t<T>> h_diag_gpu(M * batch_count);
|
|
CHECK_HIP_ERROR(hipMemcpy(h_diag_gpu,
|
|
d_diag,
|
|
sizeof(floating_data_t<T>) * M * batch_count,
|
|
hipMemcpyDeviceToHost));
|
|
|
|
if(ROCSPARSE_REPRODUCIBILITY)
|
|
{
|
|
rocsparse_reproducibility::save(
|
|
"L values", h_csr_val_gpu, "D values", h_diag_gpu);
|
|
}
|
|
|
|
// Every batch's D must land in its own slice of the device vector.
|
|
h_diag_ref.near_check(h_diag_gpu, tol_ildlt);
|
|
}
|
|
}
|
|
}
|
|
}
|
|
|
|
if(arg.timing)
|
|
{
|
|
size_t buffer_size_in_bytes = std::numeric_limits<size_t>::max();
|
|
CHECK_ROCSPARSE_ERROR(rocsparse_spildlt0_buffer_size(handle,
|
|
spildlt0_descr,
|
|
A,
|
|
A,
|
|
rocsparse_spildlt0_stage_compute,
|
|
&buffer_size_in_bytes,
|
|
p_error));
|
|
|
|
device_dense_vector<char> buffer(buffer_size_in_bytes);
|
|
CHECK_HIP_ERROR(hipMemset(buffer, 255 - 1, buffer_size_in_bytes));
|
|
|
|
const int32_t n_cold_calls = 2;
|
|
const int32_t n_sub_calls = arg.iters_inner;
|
|
const int32_t n_calls = arg.iters;
|
|
|
|
for(int32_t iter = 0; iter < n_cold_calls; ++iter)
|
|
{
|
|
A.reinit_values();
|
|
CHECK_ROCSPARSE_ERROR(rocsparse_spildlt0(handle,
|
|
spildlt0_descr,
|
|
A,
|
|
A,
|
|
rocsparse_spildlt0_stage_compute,
|
|
buffer_size_in_bytes,
|
|
buffer,
|
|
p_error));
|
|
CHECK_HIP_ERROR(hipStreamSynchronize(stream));
|
|
}
|
|
|
|
std::vector<double> gpu_time(n_calls);
|
|
rocsparse_clients::timer t(stream);
|
|
for(int32_t iter = 0; iter < n_calls; ++iter)
|
|
{
|
|
gpu_time[iter] = 0;
|
|
for(int32_t sub_iter = 0; sub_iter < n_sub_calls; ++sub_iter)
|
|
{
|
|
A.reinit_values();
|
|
t.start();
|
|
CHECK_ROCSPARSE_ERROR(rocsparse_spildlt0(handle,
|
|
spildlt0_descr,
|
|
A,
|
|
A,
|
|
rocsparse_spildlt0_stage_compute,
|
|
buffer_size_in_bytes,
|
|
buffer,
|
|
p_error));
|
|
CHECK_HIP_ERROR(hipStreamSynchronize(stream));
|
|
const double t_microseconds = (t.stop() * 1000);
|
|
gpu_time[iter] += t_microseconds;
|
|
}
|
|
gpu_time[iter] /= n_sub_calls;
|
|
}
|
|
std::sort(gpu_time.begin(), gpu_time.end());
|
|
const int32_t mid = n_calls / 2;
|
|
const double gpu_time_used
|
|
= (n_calls % 2 == 0) ? (gpu_time[mid] + gpu_time[mid - 1]) / 2 : gpu_time[mid];
|
|
|
|
auto& device = A.template as<rocsparse_format_csr>().device();
|
|
double gbyte_count = csrildlt0_gbyte_count<T>(device.m, device.nnz);
|
|
const double gpu_gbyte = get_gpu_gbyte(gpu_time_used, gbyte_count);
|
|
display_timing_info(display_key_t::M,
|
|
device.m,
|
|
display_key_t::nnz_A,
|
|
device.nnz,
|
|
display_key_t::bandwidth,
|
|
gpu_gbyte,
|
|
display_key_t::time_ms,
|
|
get_gpu_time_msec(gpu_time_used));
|
|
}
|
|
}
|
|
|
|
void testing_spildlt0_extra(const Arguments& arg) {}
|
|
|
|
#define INSTANTIATE(I, J, T) \
|
|
template void testing_spildlt0_bad_arg<I, J, T>(const Arguments& arg); \
|
|
template void testing_spildlt0<I, J, T>(const Arguments& arg)
|
|
|
|
INSTANTIATE(int32_t, int32_t, float);
|
|
INSTANTIATE(int32_t, int32_t, double);
|
|
INSTANTIATE(int32_t, int32_t, rocsparse_float_complex);
|
|
INSTANTIATE(int32_t, int32_t, rocsparse_double_complex);
|
|
|
|
INSTANTIATE(int64_t, int32_t, float);
|
|
INSTANTIATE(int64_t, int32_t, double);
|
|
INSTANTIATE(int64_t, int32_t, rocsparse_float_complex);
|
|
INSTANTIATE(int64_t, int32_t, rocsparse_double_complex);
|
|
|
|
INSTANTIATE(int64_t, int64_t, float);
|
|
INSTANTIATE(int64_t, int64_t, double);
|
|
INSTANTIATE(int64_t, int64_t, rocsparse_float_complex);
|
|
INSTANTIATE(int64_t, int64_t, rocsparse_double_complex);
|
|
|
|
#undef INSTANTIATE
|