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
1064 lines
46 KiB
C++
1064 lines
46 KiB
C++
/* **************************************************************************
|
|
* Copyright (C) 2022-2025 Advanced Micro Devices, Inc. All rights reserved.
|
|
*
|
|
* Redistribution and use in source and binary forms, with or without
|
|
* modification, are permitted provided that the following conditions
|
|
* are met:
|
|
*
|
|
* 1. Redistributions of source code must retain the above copyright
|
|
* notice, this list of conditions and the following disclaimer.
|
|
*
|
|
* 2. Redistributions in binary form must reproduce the above copyright
|
|
* notice, this list of conditions and the following disclaimer in the
|
|
* documentation and/or other materials provided with the distribution.
|
|
*
|
|
* THIS SOFTWARE IS PROVIDED BY THE AUTHOR AND CONTRIBUTORS ``AS IS'' AND
|
|
* ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE
|
|
* IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE
|
|
* ARE DISCLAIMED. IN NO EVENT SHALL THE AUTHOR OR CONTRIBUTORS BE LIABLE
|
|
* FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL
|
|
* DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS
|
|
* OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION)
|
|
* HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT
|
|
* LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY
|
|
* OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF
|
|
* SUCH DAMAGE.
|
|
* *************************************************************************/
|
|
|
|
#pragma once
|
|
|
|
#include "common/misc/client_util.hpp"
|
|
#include "common/misc/clientcommon.hpp"
|
|
#include "common/misc/lapack_host_reference.hpp"
|
|
#include "common/misc/norm.hpp"
|
|
#include "common/misc/rocsolver.hpp"
|
|
#include "common/misc/rocsolver_arguments.hpp"
|
|
#include "common/misc/rocsolver_test.hpp"
|
|
#include "common/misc/rocsolver_timer.hpp"
|
|
|
|
template <bool STRIDED, typename T, typename S, typename W>
|
|
void gesvdx_checkBadArgs(const rocblas_handle handle,
|
|
const rocblas_svect left_svect,
|
|
const rocblas_svect right_svect,
|
|
const rocblas_srange srange,
|
|
const rocblas_int m,
|
|
const rocblas_int n,
|
|
W dA,
|
|
const rocblas_int lda,
|
|
const rocblas_stride stA,
|
|
const S vl,
|
|
const S vu,
|
|
const rocblas_int il,
|
|
const rocblas_int iu,
|
|
rocblas_int* dNsv,
|
|
S* dS,
|
|
const rocblas_stride stS,
|
|
T dU,
|
|
const rocblas_int ldu,
|
|
const rocblas_stride stU,
|
|
T dV,
|
|
const rocblas_int ldv,
|
|
const rocblas_stride stV,
|
|
rocblas_int* difail,
|
|
const rocblas_stride stF,
|
|
rocblas_int* dinfo,
|
|
const rocblas_int bc)
|
|
{
|
|
// handle
|
|
EXPECT_ROCBLAS_STATUS(rocsolver_gesvdx(STRIDED, nullptr, left_svect, right_svect, srange, m, n,
|
|
dA, lda, stA, vl, vu, il, iu, dNsv, dS, stS, dU, ldu,
|
|
stU, dV, ldv, stV, difail, stF, dinfo, bc),
|
|
rocblas_status_invalid_handle);
|
|
|
|
// values
|
|
EXPECT_ROCBLAS_STATUS(rocsolver_gesvdx(STRIDED, handle, rocblas_svect_all, right_svect, srange,
|
|
m, n, dA, lda, stA, vl, vu, il, iu, dNsv, dS, stS, dU,
|
|
ldu, stU, dV, ldv, stV, difail, stF, dinfo, bc),
|
|
rocblas_status_invalid_value);
|
|
EXPECT_ROCBLAS_STATUS(rocsolver_gesvdx(STRIDED, handle, left_svect, rocblas_svect_all, srange,
|
|
m, n, dA, lda, stA, vl, vu, il, iu, dNsv, dS, stS, dU,
|
|
ldu, stU, dV, ldv, stV, difail, stF, dinfo, bc),
|
|
rocblas_status_invalid_value);
|
|
EXPECT_ROCBLAS_STATUS(rocsolver_gesvdx(STRIDED, handle, left_svect, right_svect, rocblas_srange(0),
|
|
m, n, dA, lda, stA, vl, vu, il, iu, dNsv, dS, stS, dU,
|
|
ldu, stU, dV, ldv, stV, difail, stF, dinfo, bc),
|
|
rocblas_status_invalid_value);
|
|
|
|
// sizes (only check batch_count if applicable)
|
|
if(STRIDED)
|
|
EXPECT_ROCBLAS_STATUS(rocsolver_gesvdx(STRIDED, handle, left_svect, right_svect, srange, m,
|
|
n, dA, lda, stA, vl, vu, il, iu, dNsv, dS, stS, dU,
|
|
ldu, stU, dV, ldv, stV, difail, stF, dinfo, -1),
|
|
rocblas_status_invalid_size);
|
|
|
|
// pointers
|
|
EXPECT_ROCBLAS_STATUS(rocsolver_gesvdx(STRIDED, handle, left_svect, right_svect, srange, m, n,
|
|
(W) nullptr, lda, stA, vl, vu, il, iu, dNsv, dS, stS, dU,
|
|
ldu, stU, dV, ldv, stV, difail, stF, dinfo, bc),
|
|
rocblas_status_invalid_pointer);
|
|
EXPECT_ROCBLAS_STATUS(rocsolver_gesvdx(STRIDED, handle, left_svect, right_svect, srange, m, n,
|
|
dA, lda, stA, vl, vu, il, iu, (rocblas_int*)nullptr, dS,
|
|
stS, dU, ldu, stU, dV, ldv, stV, difail, stF, dinfo, bc),
|
|
rocblas_status_invalid_pointer);
|
|
EXPECT_ROCBLAS_STATUS(rocsolver_gesvdx(STRIDED, handle, left_svect, right_svect, srange, m, n,
|
|
dA, lda, stA, vl, vu, il, iu, dNsv, (S*)nullptr, stS, dU,
|
|
ldu, stU, dV, ldv, stV, difail, stF, dinfo, bc),
|
|
rocblas_status_invalid_pointer);
|
|
EXPECT_ROCBLAS_STATUS(rocsolver_gesvdx(STRIDED, handle, left_svect, right_svect, srange, m, n,
|
|
dA, lda, stA, vl, vu, il, iu, dNsv, dS, stS, (T) nullptr,
|
|
ldu, stU, dV, ldv, stV, difail, stF, dinfo, bc),
|
|
rocblas_status_invalid_pointer);
|
|
EXPECT_ROCBLAS_STATUS(rocsolver_gesvdx(STRIDED, handle, left_svect, right_svect, srange, m, n,
|
|
dA, lda, stA, vl, vu, il, iu, dNsv, dS, stS, dU, ldu,
|
|
stU, (T) nullptr, ldv, stV, difail, stF, dinfo, bc),
|
|
rocblas_status_invalid_pointer);
|
|
EXPECT_ROCBLAS_STATUS(rocsolver_gesvdx(STRIDED, handle, left_svect, right_svect, srange, m, n,
|
|
dA, lda, stA, vl, vu, il, iu, dNsv, dS, stS, dU, ldu,
|
|
stU, dV, ldv, stV, (rocblas_int*)nullptr, stF, dinfo, bc),
|
|
rocblas_status_invalid_pointer);
|
|
EXPECT_ROCBLAS_STATUS(rocsolver_gesvdx(STRIDED, handle, left_svect, right_svect, srange, m, n,
|
|
dA, lda, stA, vl, vu, il, iu, dNsv, dS, stS, dU, ldu, stU,
|
|
dV, ldv, stV, difail, stF, (rocblas_int*)nullptr, bc),
|
|
rocblas_status_invalid_pointer);
|
|
|
|
// quick return with invalid pointers
|
|
EXPECT_ROCBLAS_STATUS(rocsolver_gesvdx(STRIDED, handle, left_svect, right_svect, srange, 0, n,
|
|
(W) nullptr, lda, stA, vl, vu, il, iu, dNsv, (S*)nullptr,
|
|
stS, (T) nullptr, ldu, stU, dV, ldv, stV,
|
|
(rocblas_int*)nullptr, stF, dinfo, bc),
|
|
rocblas_status_success);
|
|
EXPECT_ROCBLAS_STATUS(rocsolver_gesvdx(STRIDED, handle, left_svect, right_svect, srange, m, 0,
|
|
(W) nullptr, lda, stA, vl, vu, il, iu, dNsv, (S*)nullptr,
|
|
stS, (T) nullptr, ldu, stU, (T) nullptr, ldv, stV,
|
|
(rocblas_int*)nullptr, stF, dinfo, bc),
|
|
rocblas_status_success);
|
|
|
|
// quick return with zero batch_count if applicable
|
|
if(STRIDED)
|
|
EXPECT_ROCBLAS_STATUS(rocsolver_gesvdx(STRIDED, handle, left_svect, right_svect, srange, m,
|
|
n, dA, lda, stA, vl, vu, il, iu,
|
|
(rocblas_int*)nullptr, dS, stS, dU, ldu, stU, dV,
|
|
ldv, stV, difail, stF, (rocblas_int*)nullptr, 0),
|
|
rocblas_status_success);
|
|
}
|
|
|
|
template <bool BATCHED, bool STRIDED, typename T>
|
|
void testing_gesvdx_bad_arg()
|
|
{
|
|
using S = decltype(std::real(T{}));
|
|
|
|
// safe arguments
|
|
rocblas_local_handle handle;
|
|
rocblas_svect left_svect = rocblas_svect_singular;
|
|
rocblas_svect right_svect = rocblas_svect_singular;
|
|
rocblas_srange srange = rocblas_srange_all;
|
|
rocblas_int m = 2;
|
|
rocblas_int n = 2;
|
|
rocblas_int lda = 2;
|
|
rocblas_int ldu = 2;
|
|
rocblas_int ldv = 2;
|
|
rocblas_stride stA = 2;
|
|
rocblas_stride stS = 2;
|
|
rocblas_stride stU = 2;
|
|
rocblas_stride stV = 2;
|
|
rocblas_stride stF = 2;
|
|
rocblas_int bc = 1;
|
|
S vl = 0;
|
|
S vu = 0;
|
|
rocblas_int il = 0;
|
|
rocblas_int iu = 0;
|
|
|
|
// memory allocations (all cases)
|
|
device_strided_batch_vector<S> dS(1, 1, 1, 1);
|
|
device_strided_batch_vector<T> dU(1, 1, 1, 1);
|
|
device_strided_batch_vector<T> dV(1, 1, 1, 1);
|
|
device_strided_batch_vector<rocblas_int> dNsv(1, 1, 1, 1);
|
|
device_strided_batch_vector<rocblas_int> difail(1, 1, 1, 1);
|
|
device_strided_batch_vector<rocblas_int> dinfo(1, 1, 1, 1);
|
|
CHECK_HIP_ERROR(dS.memcheck());
|
|
CHECK_HIP_ERROR(dU.memcheck());
|
|
CHECK_HIP_ERROR(dV.memcheck());
|
|
CHECK_HIP_ERROR(dNsv.memcheck());
|
|
CHECK_HIP_ERROR(difail.memcheck());
|
|
CHECK_HIP_ERROR(dinfo.memcheck());
|
|
|
|
if(BATCHED)
|
|
{
|
|
// memory allocations
|
|
device_batch_vector<T> dA(1, 1, 1);
|
|
CHECK_HIP_ERROR(dA.memcheck());
|
|
|
|
// check bad arguments
|
|
gesvdx_checkBadArgs<STRIDED>(handle, left_svect, right_svect, srange, m, n, dA.data(), lda,
|
|
stA, vl, vu, il, iu, dNsv, dS.data(), stS, dU.data(), ldu, stU,
|
|
dV.data(), ldv, stV, difail.data(), stF, dinfo.data(), bc);
|
|
}
|
|
else
|
|
{
|
|
// memory allocations
|
|
device_strided_batch_vector<T> dA(1, 1, 1, 1);
|
|
CHECK_HIP_ERROR(dA.memcheck());
|
|
|
|
// check bad arguments
|
|
gesvdx_checkBadArgs<STRIDED>(handle, left_svect, right_svect, srange, m, n, dA.data(), lda,
|
|
stA, vl, vu, il, iu, dNsv, dS.data(), stS, dU.data(), ldu, stU,
|
|
dV.data(), ldv, stV, difail.data(), stF, dinfo.data(), bc);
|
|
}
|
|
}
|
|
|
|
template <bool CPU, bool GPU, typename T, typename Td, typename Th>
|
|
void gesvdx_initData(const rocblas_handle handle,
|
|
const rocblas_svect left_svect,
|
|
const rocblas_svect right_svect,
|
|
const rocblas_int m,
|
|
const rocblas_int n,
|
|
Td& dA,
|
|
const rocblas_int lda,
|
|
const rocblas_int bc,
|
|
Th& hA,
|
|
std::vector<T>& A,
|
|
bool test = true)
|
|
{
|
|
if(CPU)
|
|
{
|
|
rocblas_init<T>(hA, true);
|
|
rocblas_int nn = std::min(m, n);
|
|
|
|
// construct non singular matrix A such that all singular values are in (0, 20]
|
|
for(rocblas_int b = 0; b < bc; ++b)
|
|
{
|
|
for(rocblas_int i = 0; i < m; i++)
|
|
{
|
|
if(i == nn / 4 || i == nn / 2 || i == nn - 1 || i == nn / 7 || i == nn / 5
|
|
|| i == nn / 3)
|
|
hA[b][i + i * lda] = 0;
|
|
|
|
for(rocblas_int j = 0; j < n; j++)
|
|
{
|
|
if(i == j)
|
|
hA[b][i + j * lda] = 2 * std::real(hA[b][i + j * lda]) - 21;
|
|
else
|
|
{
|
|
if(m >= n)
|
|
{
|
|
if(j == i + 1)
|
|
hA[b][i + j * lda] = (hA[b][i + j * lda] - 5) / 10;
|
|
else
|
|
hA[b][i + j * lda] = 0;
|
|
}
|
|
else
|
|
{
|
|
if(i == j + 1)
|
|
hA[b][i + j * lda] = (hA[b][i + j * lda] - 5) / 10;
|
|
else
|
|
hA[b][i + j * lda] = 0;
|
|
}
|
|
}
|
|
}
|
|
}
|
|
|
|
// make copy of original data to test vectors if required
|
|
if(test && (left_svect != rocblas_svect_none || right_svect != rocblas_svect_none))
|
|
{
|
|
for(rocblas_int i = 0; i < m; i++)
|
|
{
|
|
for(rocblas_int j = 0; j < n; j++)
|
|
A[b * lda * n + i + j * lda] = hA[b][i + j * lda];
|
|
}
|
|
}
|
|
}
|
|
}
|
|
|
|
if(GPU)
|
|
{
|
|
// now copy to the GPU
|
|
CHECK_HIP_ERROR(dA.transfer_from(hA));
|
|
}
|
|
}
|
|
|
|
template <bool STRIDED,
|
|
typename T,
|
|
typename S,
|
|
typename Wd,
|
|
typename Td,
|
|
typename Ud,
|
|
typename Id,
|
|
typename Wh,
|
|
typename Th,
|
|
typename Uh,
|
|
typename Ih>
|
|
void gesvdx_getError(const rocblas_handle handle,
|
|
const rocblas_svect left_svect,
|
|
const rocblas_svect right_svect,
|
|
const rocblas_srange srange,
|
|
const rocblas_int m,
|
|
const rocblas_int n,
|
|
Wd& dA,
|
|
const rocblas_int lda,
|
|
const rocblas_stride stA,
|
|
const S vl,
|
|
const S vu,
|
|
const rocblas_int il,
|
|
const rocblas_int iu,
|
|
Id& dNsv,
|
|
Ud& dS,
|
|
const rocblas_stride stS,
|
|
Td& dU,
|
|
const rocblas_int ldu,
|
|
const rocblas_stride stU,
|
|
Td& dV,
|
|
const rocblas_int ldv,
|
|
const rocblas_stride stV,
|
|
Id& difail,
|
|
const rocblas_stride stF,
|
|
Id& dinfo,
|
|
const rocblas_int bc,
|
|
const rocblas_svect left_svectT,
|
|
const rocblas_svect right_svectT,
|
|
const rocblas_int mT,
|
|
const rocblas_int nT,
|
|
Td& dUT,
|
|
const rocblas_int lduT,
|
|
const rocblas_stride stUT,
|
|
Td& dVT,
|
|
const rocblas_int ldvT,
|
|
const rocblas_stride stVT,
|
|
Wh& hA,
|
|
Ih& hNsv,
|
|
Ih& hNsvRes,
|
|
Uh& hS,
|
|
Uh& hSres,
|
|
Th& hU,
|
|
Th& hUres,
|
|
const rocblas_int ldures,
|
|
Th& hV,
|
|
Th& hVres,
|
|
const rocblas_int ldvres,
|
|
Ih& hifail,
|
|
Ih& hifailRes,
|
|
Ih& hinfo,
|
|
Ih& hinfoRes,
|
|
double* max_err,
|
|
double* max_errv)
|
|
{
|
|
/** As per lapack's documentation, the following workspace size should work:
|
|
rocblas_int minn = std::min(m,n);
|
|
rocblas_int maxn = std::max(m,n);
|
|
rocblas_int lwork = minn * minn + 6 * minn + maxn;
|
|
rocblas_int lrwork = 17 * minn * minn;
|
|
std::vector<T> work(lwork);
|
|
std::vector<S> rwork(lrwork);
|
|
HOWEVER, gesvdx_ GIVES ILLEGAL VALUE FOR ARGUMENT lwork.
|
|
|
|
Making the memory query to get the correct workspace dimension:
|
|
std::vector<T> query(1);
|
|
cpu_gesvdx(left_svect, right_svect, srange, m, n, hA[0], lda, vl, vu, il, iu, hNsv[0], hS[0], hU[0], ldu, hV[0], ldv,
|
|
query.data(), -1, rwork.data(), hifail[0], hinfo[0]);
|
|
rocblas_int lwork = int(std::real(query[0]));
|
|
std::vector<T> work(lwork);
|
|
AND NOW gesvdx_ FAILS WITH seg fault ERROR. **/
|
|
|
|
// (TODO: Need to confirm problem with gesvdx_ and report it)
|
|
|
|
/** WORKAROUND: for now, we will call gesvd_ to get all the singular values on the CPU side and
|
|
offset the result array according to srange, vl, vu, il, and iu. This approach has 2 disadvantages:
|
|
1. singular values are not computed to the same accuracy by gesvd_ (QR iteration) and
|
|
gesvdx_ (inverse iteration). So, comparison maybe more sensitive.
|
|
2. info and ifail cannot be tested as they have different meaning in gesvd_
|
|
3. we cannot provide timing for CPU execution using gesvd_ when testing gesvdx_ **/
|
|
|
|
// (TODO: We may revisit the entire approach in the future: change to another solution,
|
|
// or wait for problems with gesvdx_ to be fixed)
|
|
|
|
std::vector<rocblas_int> offset(bc);
|
|
rocblas_int lwork = 5 * std::max(m, n);
|
|
rocblas_int lrwork = (rocblas_is_complex<T> ? 5 * std::min(m, n) : 0);
|
|
std::vector<T> work(lwork);
|
|
std::vector<S> rwork(lrwork);
|
|
rocblas_int minn = std::min(m, n);
|
|
|
|
// input data initialization
|
|
std::vector<T> A(lda * n * bc);
|
|
gesvdx_initData<true, true, T>(handle, left_svect, right_svect, m, n, dA, lda, bc, hA, A);
|
|
|
|
// execute computations:
|
|
// complementary execution to compute all singular vectors if needed
|
|
if(mT * nT > 0)
|
|
{
|
|
CHECK_ROCBLAS_ERROR(rocsolver_gesvdx(STRIDED, handle, left_svectT, right_svectT, srange, mT,
|
|
nT, dA.data(), lda, stA, vl, vu, il, iu, dNsv.data(),
|
|
dS.data(), stS, dUT.data(), lduT, stUT, dVT.data(),
|
|
ldvT, stVT, difail.data(), stF, dinfo.data(), bc));
|
|
|
|
if(left_svect == rocblas_svect_none && right_svect != rocblas_svect_none)
|
|
CHECK_HIP_ERROR(hUres.transfer_from(dUT));
|
|
if(right_svect == rocblas_svect_none && left_svect != rocblas_svect_none)
|
|
CHECK_HIP_ERROR(hVres.transfer_from(dVT));
|
|
}
|
|
|
|
gesvdx_initData<false, true, T>(handle, left_svect, right_svect, m, n, dA, lda, bc, hA, A);
|
|
|
|
// CPU lapack
|
|
for(rocblas_int b = 0; b < bc; ++b)
|
|
{
|
|
//cpu_gesvdx(rocblas_svect_none, rocblas_svect_none, srange, m, n, hA[b], lda, vl, vu, il, iu, hNsv[b], hS[b], hU[b], ldu, hV[b], ldv,
|
|
// work.data(), lwork, rwork.data(), hifail[b], hinfo[b]);
|
|
|
|
/*** WORKAROUND: ***/
|
|
cpu_gesvd(rocblas_svect_none, rocblas_svect_none, m, n, hA[b], lda, hS[b], hU[b], ldu,
|
|
hV[b], ldv, work.data(), lwork, rwork.data(), hinfo[b]);
|
|
hNsv[b][0] = 0;
|
|
offset[b] = -1;
|
|
if(srange == rocblas_srange_index)
|
|
{
|
|
offset[b] = il - 1;
|
|
hNsv[b][0] = iu - il + 1;
|
|
}
|
|
else if(srange == rocblas_srange_value)
|
|
{
|
|
for(int j = 0; j < minn; ++j)
|
|
{
|
|
if(hS[b][j] < vu && hS[b][j] >= vl)
|
|
{
|
|
if(offset[b] == -1)
|
|
offset[b] = j;
|
|
hNsv[b][0]++;
|
|
}
|
|
}
|
|
}
|
|
else
|
|
{
|
|
offset[b] = 0;
|
|
hNsv[b][0] = minn;
|
|
}
|
|
/*******************/
|
|
}
|
|
|
|
// GPU lapack
|
|
CHECK_ROCBLAS_ERROR(rocsolver_gesvdx(STRIDED, handle, left_svect, right_svect, srange, m, n,
|
|
dA.data(), lda, stA, vl, vu, il, iu, dNsv.data(),
|
|
dS.data(), stS, dU.data(), ldu, stU, dV.data(), ldv, stV,
|
|
difail.data(), stF, dinfo.data(), bc));
|
|
|
|
CHECK_HIP_ERROR(hSres.transfer_from(dS));
|
|
CHECK_HIP_ERROR(hNsvRes.transfer_from(dNsv));
|
|
CHECK_HIP_ERROR(hifailRes.transfer_from(difail));
|
|
CHECK_HIP_ERROR(hinfoRes.transfer_from(dinfo));
|
|
|
|
if(left_svect == rocblas_svect_singular)
|
|
CHECK_HIP_ERROR(hUres.transfer_from(dU));
|
|
if(right_svect == rocblas_svect_singular)
|
|
CHECK_HIP_ERROR(hVres.transfer_from(dV));
|
|
|
|
*max_err = 0;
|
|
*max_errv = 0;
|
|
|
|
// Check info and ifail for non-convergence
|
|
// (NOTE: With the workaround in place, info and ifail cannot be tested as they have different
|
|
// meaning in gesvd_, however, We expect the used input matrices to always converge)
|
|
/*for(rocblas_int b = 0; b < bc; ++b)
|
|
{
|
|
EXPECT_EQ(hinfo[b][0], hinfoRes[b][0]) << "where b = " << b;
|
|
if(hinfo[b][0] != hinfoRes[b][0])
|
|
*max_err += 1;
|
|
for(int j = 0; j < hNsv[b][0]; ++j)
|
|
{
|
|
EXPECT_EQ(hifail[b][j], hifailRes[b][j]) << "where b = " << b << ", j = " << j;
|
|
if(hifail[b][j] != hifailRes[b][j])
|
|
*max_err += 1;
|
|
}
|
|
}*/
|
|
|
|
double err = 0;
|
|
for(rocblas_int b = 0; b < bc; ++b)
|
|
{
|
|
// check number of computed singular values
|
|
rocblas_int nn = hNsvRes[b][0];
|
|
*max_err += std::abs(nn - hNsv[b][0]);
|
|
EXPECT_EQ(hNsv[b][0], hNsvRes[b][0]) << "where b = " << b;
|
|
|
|
// error is ||hS - hSres||
|
|
err = norm_error('F', 1, hNsv[b][0], 1, hS[b] + offset[b], hSres[b]); //WORKAROUND
|
|
*max_err = err > *max_err ? err : *max_err;
|
|
|
|
// Check the singular vectors if required
|
|
if(hinfo[b][0] == 0 && (left_svect != rocblas_svect_none || right_svect != rocblas_svect_none))
|
|
{
|
|
// U and V should be orthonormal, if they are then U^T*U and V*V^T should be the identity
|
|
if(nn > 0)
|
|
{
|
|
std::vector<T> UUres(nn * nn, 0.0);
|
|
std::vector<T> VVres(nn * nn, 0.0);
|
|
std::vector<T> I(nn * nn, 0.0);
|
|
|
|
for(rocblas_int i = 0; i < nn; i++)
|
|
I[i + i * nn] = T(1);
|
|
|
|
cpu_gemm(rocblas_operation_conjugate_transpose, rocblas_operation_none, nn, nn, m,
|
|
T(1), hUres[b], ldures, hUres[b], ldures, T(0), UUres.data(), nn);
|
|
err = norm_error('F', nn, nn, nn, I.data(), UUres.data());
|
|
*max_errv = err > *max_errv ? err : *max_errv;
|
|
|
|
cpu_gemm(rocblas_operation_none, rocblas_operation_conjugate_transpose, nn, nn, n,
|
|
T(1), hVres[b], ldvres, hVres[b], ldvres, T(0), VVres.data(), nn);
|
|
err = norm_error('F', nn, nn, nn, I.data(), VVres.data());
|
|
*max_errv = err > *max_errv ? err : *max_errv;
|
|
}
|
|
|
|
err = 0;
|
|
// check singular vectors implicitly (A*v_k = s_k*u_k)
|
|
for(rocblas_int k = 0; k < hNsv[b][0]; ++k)
|
|
{
|
|
T tmp = 0;
|
|
double tmp2 = 0;
|
|
|
|
// (Comparing absolute values to deal with the fact that the pair of singular vectors (u,-v) or (-u,v) are
|
|
// both ok and we could get either one with the complementary or main executions when only
|
|
// one side set of vectors is required. May be revisited in the future.)
|
|
for(rocblas_int i = 0; i < m; ++i)
|
|
{
|
|
tmp = 0;
|
|
for(rocblas_int j = 0; j < n; ++j)
|
|
tmp += A[b * lda * n + i + j * lda] * sconj(hVres[b][k + j * ldvres]);
|
|
tmp2 = std::abs(tmp) - std::abs(hSres[b][k] * hUres[b][i + k * ldures]);
|
|
err += tmp2 * tmp2;
|
|
}
|
|
}
|
|
err = std::sqrt(err) / double(snorm('F', m, n, A.data() + b * lda * n, lda));
|
|
*max_errv = err > *max_errv ? err : *max_errv;
|
|
}
|
|
}
|
|
}
|
|
|
|
template <bool STRIDED,
|
|
typename T,
|
|
typename S,
|
|
typename Wd,
|
|
typename Td,
|
|
typename Ud,
|
|
typename Id,
|
|
typename Wh,
|
|
typename Th,
|
|
typename Uh,
|
|
typename Ih>
|
|
void gesvdx_getPerfData(const rocblas_handle handle,
|
|
const rocblas_svect left_svect,
|
|
const rocblas_svect right_svect,
|
|
const rocblas_srange srange,
|
|
const rocblas_int m,
|
|
const rocblas_int n,
|
|
Wd& dA,
|
|
const rocblas_int lda,
|
|
const rocblas_stride stA,
|
|
const S vl,
|
|
const S vu,
|
|
const rocblas_int il,
|
|
const rocblas_int iu,
|
|
Id& dNsv,
|
|
Ud& dS,
|
|
const rocblas_stride stS,
|
|
Td& dU,
|
|
const rocblas_int ldu,
|
|
const rocblas_stride stU,
|
|
Td& dV,
|
|
const rocblas_int ldv,
|
|
const rocblas_stride stV,
|
|
Id& difail,
|
|
const rocblas_stride stF,
|
|
Id& dinfo,
|
|
const rocblas_int bc,
|
|
Wh& hA,
|
|
Ih& hNsv,
|
|
Uh& hS,
|
|
Th& hU,
|
|
Th& hV,
|
|
Ih& hifail,
|
|
Ih& hinfo,
|
|
double* gpu_time_used,
|
|
double* cpu_time_used,
|
|
const rocblas_int hot_calls,
|
|
const int profile,
|
|
const bool profile_kernels,
|
|
const bool perf)
|
|
{
|
|
/** As per lapack's documentation, the following workspace size should work:
|
|
rocblas_int minn = std::min(m,n);
|
|
rocblas_int maxn = std::max(m,n);
|
|
rocblas_int lwork = minn * minn + 6 * minn + maxn;
|
|
rocblas_int lrwork = 17 * minn * minn;
|
|
std::vector<T> work(lwork);
|
|
std::vector<S> rwork(lrwork);
|
|
HOWEVER, gesvdx_ GIVES ILLEGAL VALUE FOR ARGUMENT lwork.
|
|
|
|
Making the memory query to get the correct workspace dimension:
|
|
std::vector<T> query(1);
|
|
cpu_gesvdx(left_svect, right_svect, srange, m, n, hA[0], lda, vl, vu, il, iu, hNsv[0], hS[0], hU[0], ldu, hV[0], ldv,
|
|
query.data(), -1, rwork.data(), hifail[0], hinfo[0]);
|
|
rocblas_int lwork = int(std::real(query[0]));
|
|
std::vector<T> work(lwork);
|
|
AND NOW gesvdx_ FAILS WITH seg fault ERROR. **/
|
|
|
|
// (TODO: Need to confirm problem with gesvdx_ and report it)
|
|
|
|
// For now we cannot report cpu time
|
|
|
|
std::vector<T> A(lda * n * bc);
|
|
|
|
if(!perf)
|
|
{
|
|
//gesvdx_initData<true, false, T>(handle, left_svect, right_svect, m, n, dA, lda, bc, hA, A, 0);
|
|
|
|
// cpu-lapack performance (only if not in perf mode)
|
|
//*cpu_time_used = get_time_us_no_sync();
|
|
//for(rocblas_int b = 0; b < bc; ++b)
|
|
// cpu_gesvdx(left_svect, right_svect, srange, m, n, hA[b], lda, vl, vu, il, iu, hNsv[b], hS[b], hU[b], ldu, hV[b], ldv,
|
|
// work.data(), lwork, rwork.data(), hifail[b], hinfo[b]);
|
|
//*cpu_time_used = get_time_us_no_sync() - *cpu_time_used;
|
|
*cpu_time_used = nan("");
|
|
}
|
|
|
|
gesvdx_initData<true, false, T>(handle, left_svect, right_svect, m, n, dA, lda, bc, hA, A, 0);
|
|
|
|
// cold calls
|
|
for(int iter = 0; iter < 2; iter++)
|
|
{
|
|
gesvdx_initData<false, true, T>(handle, left_svect, right_svect, m, n, dA, lda, bc, hA, A, 0);
|
|
|
|
CHECK_ROCBLAS_ERROR(rocsolver_gesvdx(STRIDED, handle, left_svect, right_svect, srange, m, n,
|
|
dA.data(), lda, stA, vl, vu, il, iu, dNsv.data(),
|
|
dS.data(), stS, dU.data(), ldu, stU, dV.data(), ldv,
|
|
stV, difail.data(), stF, dinfo.data(), bc));
|
|
}
|
|
|
|
// gpu-lapack performance
|
|
hipStream_t stream;
|
|
CHECK_ROCBLAS_ERROR(rocblas_get_stream(handle, &stream));
|
|
rocsolver_timer timer;
|
|
|
|
if(profile > 0)
|
|
{
|
|
if(profile_kernels)
|
|
rocsolver_log_set_layer_mode(rocblas_layer_mode_log_profile
|
|
| rocblas_layer_mode_ex_log_kernel);
|
|
else
|
|
rocsolver_log_set_layer_mode(rocblas_layer_mode_log_profile);
|
|
rocsolver_log_set_max_levels(profile);
|
|
}
|
|
|
|
for(rocblas_int iter = 0; iter < hot_calls; iter++)
|
|
{
|
|
gesvdx_initData<false, true, T>(handle, left_svect, right_svect, m, n, dA, lda, bc, hA, A, 0);
|
|
|
|
timer.start(stream);
|
|
rocsolver_gesvdx(STRIDED, handle, left_svect, right_svect, srange, m, n, dA.data(), lda,
|
|
stA, vl, vu, il, iu, dNsv.data(), dS.data(), stS, dU.data(), ldu, stU,
|
|
dV.data(), ldv, stV, difail.data(), stF, dinfo.data(), bc);
|
|
timer.end(stream);
|
|
}
|
|
*gpu_time_used = timer.get_combined();
|
|
}
|
|
|
|
template <bool BATCHED, bool STRIDED, typename T>
|
|
void testing_gesvdx(Arguments& argus)
|
|
{
|
|
using S = decltype(std::real(T{}));
|
|
|
|
// get arguments
|
|
rocblas_local_handle handle;
|
|
char leftvC = argus.get<char>("left_svect");
|
|
char rightvC = argus.get<char>("right_svect");
|
|
char srangeC = argus.get<char>("srange");
|
|
rocblas_svect leftv = char2rocblas_svect(leftvC);
|
|
rocblas_svect rightv = char2rocblas_svect(rightvC);
|
|
rocblas_srange srange = char2rocblas_srange(srangeC);
|
|
|
|
S vl = S(argus.get<double>("vl", 0));
|
|
S vu = S(argus.get<double>("vu", srangeC == 'V' ? 1 : 0));
|
|
rocblas_int il = argus.get<rocblas_int>("il", srangeC == 'I' ? 1 : 0);
|
|
rocblas_int iu = argus.get<rocblas_int>("iu", srangeC == 'I' ? 1 : 0);
|
|
|
|
rocblas_int m = argus.get<rocblas_int>("m");
|
|
rocblas_int n = argus.get<rocblas_int>("n", m);
|
|
rocblas_int nn = std::min(m, n);
|
|
rocblas_int nsv_max = (srange == rocblas_srange_index ? iu - il + 1 : nn);
|
|
rocblas_int lda = argus.get<rocblas_int>("lda", m);
|
|
rocblas_int ldu = argus.get<rocblas_int>("ldu", m);
|
|
rocblas_int ldv = argus.get<rocblas_int>("ldv", nsv_max);
|
|
rocblas_stride stA = argus.get<rocblas_stride>("strideA", lda * n);
|
|
rocblas_stride stS = argus.get<rocblas_stride>("strideS", nsv_max);
|
|
rocblas_stride stF = argus.get<rocblas_stride>("strideF", nn);
|
|
rocblas_stride stU = argus.get<rocblas_stride>("strideU", ldu * nsv_max);
|
|
rocblas_stride stV = argus.get<rocblas_stride>("strideV", ldv * n);
|
|
|
|
rocblas_int bc = argus.batch_count;
|
|
rocblas_int hot_calls = argus.iters;
|
|
|
|
// check non-supported values
|
|
if(rightv == rocblas_svect_overwrite || leftv == rocblas_svect_overwrite
|
|
|| rightv == rocblas_svect_all || leftv == rocblas_svect_all)
|
|
{
|
|
if(BATCHED)
|
|
EXPECT_ROCBLAS_STATUS(
|
|
rocsolver_gesvdx(STRIDED, handle, leftv, rightv, srange, m, n, (T* const*)nullptr,
|
|
lda, stA, vl, vu, il, iu, (rocblas_int*)nullptr, (S*)nullptr, stS,
|
|
(T*)nullptr, ldu, stU, (T*)nullptr, ldv, stV,
|
|
(rocblas_int*)nullptr, stF, (rocblas_int*)nullptr, bc),
|
|
rocblas_status_invalid_value);
|
|
else
|
|
EXPECT_ROCBLAS_STATUS(
|
|
rocsolver_gesvdx(STRIDED, handle, leftv, rightv, srange, m, n, (T*)nullptr, lda,
|
|
stA, vl, vu, il, iu, (rocblas_int*)nullptr, (S*)nullptr, stS,
|
|
(T*)nullptr, ldu, stU, (T*)nullptr, ldv, stV,
|
|
(rocblas_int*)nullptr, stF, (rocblas_int*)nullptr, bc),
|
|
rocblas_status_invalid_value);
|
|
|
|
if(argus.timing)
|
|
rocsolver_bench_inform(inform_invalid_args);
|
|
|
|
return;
|
|
}
|
|
|
|
/** TESTING OF SINGULAR VECTORS IS DONE IMPLICITLY, NOT EXPLICITLY COMPARING
|
|
WITH LAPACK. SO, WE ALWAYS NEED TO COMPUTE THE SAME NUMBER OF ELEMENTS OF
|
|
THE RIGHT AND LEFT VECTORS. WHILE DOING THIS, IF MORE VECTORS THAN THE
|
|
SPECIFIED IN THE MAIN CALL NEED TO BE COMPUTED, WE DO SO WITH AN EXTRA CALL **/
|
|
|
|
rocblas_svect leftvT = rocblas_svect_none;
|
|
rocblas_svect rightvT = rocblas_svect_none;
|
|
rocblas_int ldvT = 1;
|
|
rocblas_int lduT = 1;
|
|
rocblas_int mT = 0;
|
|
rocblas_int nT = 0;
|
|
bool svects = (leftv != rocblas_svect_none || rightv != rocblas_svect_none);
|
|
|
|
if(svects)
|
|
{
|
|
if(leftv == rocblas_svect_none)
|
|
{
|
|
leftvT = rocblas_svect_singular;
|
|
lduT = m;
|
|
mT = m;
|
|
nT = n;
|
|
}
|
|
if(rightv == rocblas_svect_none)
|
|
{
|
|
rightvT = rocblas_svect_singular;
|
|
ldvT = nsv_max;
|
|
mT = m;
|
|
nT = n;
|
|
}
|
|
}
|
|
|
|
// determine sizes
|
|
rocblas_int ldures = 1;
|
|
rocblas_int ldvres = 1;
|
|
size_t size_hSres = 0;
|
|
size_t size_hUres = 0;
|
|
size_t size_hVres = 0;
|
|
size_t size_hifailRes = 0;
|
|
size_t size_UT = 0;
|
|
size_t size_VT = 0;
|
|
size_t size_A = size_t(lda) * n;
|
|
size_t size_S = size_t(nsv_max);
|
|
size_t size_S_cpu = size_t(nn);
|
|
size_t size_V = size_t(ldv) * n;
|
|
size_t size_U = size_t(ldu) * nsv_max;
|
|
size_t size_ifail = nn;
|
|
if(argus.unit_check || argus.norm_check)
|
|
{
|
|
size_hifailRes = nn;
|
|
size_VT = size_t(ldvT) * n;
|
|
size_UT = size_t(lduT) * nsv_max;
|
|
size_hSres = nsv_max;
|
|
if(svects)
|
|
{
|
|
if(leftv == rocblas_svect_none)
|
|
{
|
|
size_hUres = size_UT;
|
|
ldures = lduT;
|
|
}
|
|
else
|
|
{
|
|
size_hUres = size_U;
|
|
ldures = ldu;
|
|
}
|
|
|
|
if(rightv == rocblas_svect_none)
|
|
{
|
|
size_hVres = size_VT;
|
|
ldvres = ldvT;
|
|
}
|
|
else
|
|
{
|
|
size_hVres = size_V;
|
|
ldvres = ldv;
|
|
}
|
|
}
|
|
}
|
|
rocblas_stride stS_cpu = size_S_cpu;
|
|
rocblas_stride stUT = size_UT;
|
|
rocblas_stride stVT = size_VT;
|
|
rocblas_stride stUres = size_hUres;
|
|
rocblas_stride stVres = size_hVres;
|
|
|
|
double max_error = 0, gpu_time_used = 0, cpu_time_used = 0, max_errorv = 0;
|
|
|
|
// check invalid sizes
|
|
bool invalid_size = (n < 0 || m < 0 || lda < m || ldu < 1 || ldv < 1 || bc < 0)
|
|
|| (leftv == rocblas_svect_singular && ldu < m)
|
|
|| (rightv == rocblas_svect_singular && ldv < nsv_max)
|
|
|| (srange == rocblas_srange_value && (vl < 0 || vl >= vu))
|
|
|| (srange == rocblas_srange_index && (il < 1 || iu < 0))
|
|
|| (srange == rocblas_srange_index && (iu > nn || (nn > 0 && il > iu)));
|
|
|
|
if(invalid_size)
|
|
{
|
|
if(BATCHED)
|
|
EXPECT_ROCBLAS_STATUS(
|
|
rocsolver_gesvdx(STRIDED, handle, leftv, rightv, srange, m, n, (T* const*)nullptr,
|
|
lda, stA, vl, vu, il, iu, (rocblas_int*)nullptr, (S*)nullptr, stS,
|
|
(T*)nullptr, ldu, stU, (T*)nullptr, ldv, stV,
|
|
(rocblas_int*)nullptr, stF, (rocblas_int*)nullptr, bc),
|
|
rocblas_status_invalid_size);
|
|
else
|
|
EXPECT_ROCBLAS_STATUS(
|
|
rocsolver_gesvdx(STRIDED, handle, leftv, rightv, srange, m, n, (T*)nullptr, lda,
|
|
stA, vl, vu, il, iu, (rocblas_int*)nullptr, (S*)nullptr, stS,
|
|
(T*)nullptr, ldu, stU, (T*)nullptr, ldv, stV,
|
|
(rocblas_int*)nullptr, stF, (rocblas_int*)nullptr, bc),
|
|
rocblas_status_invalid_size);
|
|
|
|
if(argus.timing)
|
|
rocsolver_bench_inform(inform_invalid_size);
|
|
|
|
return;
|
|
}
|
|
|
|
// memory size query is necessary
|
|
if(argus.mem_query)
|
|
{
|
|
CHECK_ROCBLAS_ERROR(rocblas_start_device_memory_size_query(handle));
|
|
if(BATCHED)
|
|
{
|
|
CHECK_ALLOC_QUERY(rocsolver_gesvdx(
|
|
STRIDED, handle, leftv, rightv, srange, m, n, (T* const*)nullptr, lda, stA, vl, vu,
|
|
il, iu, (rocblas_int*)nullptr, (S*)nullptr, stS, (T*)nullptr, ldu, stU, (T*)nullptr,
|
|
ldv, stV, (rocblas_int*)nullptr, stF, (rocblas_int*)nullptr, bc));
|
|
CHECK_ALLOC_QUERY(rocsolver_gesvdx(
|
|
STRIDED, handle, leftvT, rightvT, srange, mT, nT, (T* const*)nullptr, lda, stA, vl,
|
|
vu, il, iu, (rocblas_int*)nullptr, (S*)nullptr, stS, (T*)nullptr, lduT, stUT,
|
|
(T*)nullptr, ldvT, stVT, (rocblas_int*)nullptr, stF, (rocblas_int*)nullptr, bc));
|
|
}
|
|
else
|
|
{
|
|
CHECK_ALLOC_QUERY(rocsolver_gesvdx(
|
|
STRIDED, handle, leftv, rightv, srange, m, n, (T*)nullptr, lda, stA, vl, vu, il, iu,
|
|
(rocblas_int*)nullptr, (S*)nullptr, stS, (T*)nullptr, ldu, stU, (T*)nullptr, ldv,
|
|
stV, (rocblas_int*)nullptr, stF, (rocblas_int*)nullptr, bc));
|
|
CHECK_ALLOC_QUERY(rocsolver_gesvdx(
|
|
STRIDED, handle, leftvT, rightvT, srange, mT, nT, (T*)nullptr, lda, stA, vl, vu, il,
|
|
iu, (rocblas_int*)nullptr, (S*)nullptr, stS, (T*)nullptr, lduT, stUT, (T*)nullptr,
|
|
ldvT, stVT, (rocblas_int*)nullptr, stF, (rocblas_int*)nullptr, bc));
|
|
}
|
|
|
|
size_t size;
|
|
CHECK_ROCBLAS_ERROR(rocblas_stop_device_memory_size_query(handle, &size));
|
|
|
|
rocsolver_bench_inform(inform_mem_query, size);
|
|
return;
|
|
}
|
|
|
|
// memory allocations (all cases)
|
|
// host
|
|
host_strided_batch_vector<S> hS(size_S_cpu, 1, stS_cpu, bc); // extra space for cpu_gesvd call
|
|
host_strided_batch_vector<T> hV(size_V, 1, stV, bc);
|
|
host_strided_batch_vector<T> hU(size_U, 1, stU, bc);
|
|
host_strided_batch_vector<rocblas_int> hNsv(1, 1, 1, bc);
|
|
host_strided_batch_vector<rocblas_int> hNsvRes(1, 1, 1, bc);
|
|
host_strided_batch_vector<rocblas_int> hifail(12 * nn, 1, stF, bc);
|
|
host_strided_batch_vector<rocblas_int> hifailRes(size_hifailRes, 1, stF, bc);
|
|
host_strided_batch_vector<rocblas_int> hinfo(1, 1, 1, bc);
|
|
host_strided_batch_vector<rocblas_int> hinfoRes(1, 1, 1, bc);
|
|
host_strided_batch_vector<S> hSres(size_hSres, 1, stS, bc);
|
|
host_strided_batch_vector<T> hVres(size_hVres, 1, stVres, bc);
|
|
host_strided_batch_vector<T> hUres(size_hUres, 1, stUres, bc);
|
|
// device
|
|
device_strided_batch_vector<S> dS(size_S, 1, stS, bc);
|
|
device_strided_batch_vector<T> dV(size_V, 1, stV, bc);
|
|
device_strided_batch_vector<T> dU(size_U, 1, stU, bc);
|
|
device_strided_batch_vector<rocblas_int> dinfo(1, 1, 1, bc);
|
|
device_strided_batch_vector<rocblas_int> dNsv(1, 1, 1, bc);
|
|
device_strided_batch_vector<rocblas_int> difail(size_ifail, 1, stF, bc);
|
|
device_strided_batch_vector<T> dVT(size_VT, 1, stVT, bc);
|
|
device_strided_batch_vector<T> dUT(size_UT, 1, stUT, bc);
|
|
if(size_VT)
|
|
CHECK_HIP_ERROR(dVT.memcheck());
|
|
if(size_UT)
|
|
CHECK_HIP_ERROR(dUT.memcheck());
|
|
if(size_S)
|
|
CHECK_HIP_ERROR(dS.memcheck());
|
|
if(size_V)
|
|
CHECK_HIP_ERROR(dV.memcheck());
|
|
if(size_U)
|
|
CHECK_HIP_ERROR(dU.memcheck());
|
|
if(size_ifail)
|
|
CHECK_HIP_ERROR(difail.memcheck());
|
|
CHECK_HIP_ERROR(dinfo.memcheck());
|
|
CHECK_HIP_ERROR(dNsv.memcheck());
|
|
|
|
if(BATCHED)
|
|
{
|
|
// memory allocations
|
|
host_batch_vector<T> hA(size_A, 1, bc);
|
|
device_batch_vector<T> dA(size_A, 1, bc);
|
|
if(size_A)
|
|
CHECK_HIP_ERROR(dA.memcheck());
|
|
|
|
// check quick return
|
|
if(n == 0 || m == 0 || bc == 0)
|
|
{
|
|
EXPECT_ROCBLAS_STATUS(rocsolver_gesvdx(STRIDED, handle, leftv, rightv, srange, m, n,
|
|
dA.data(), lda, stA, vl, vu, il, iu, dNsv.data(),
|
|
dS.data(), stS, dU.data(), ldu, stU, dV.data(),
|
|
ldv, stV, difail.data(), stF, dinfo.data(), bc),
|
|
rocblas_status_success);
|
|
if(argus.timing)
|
|
rocsolver_bench_inform(inform_quick_return);
|
|
|
|
return;
|
|
}
|
|
|
|
// check computations
|
|
if(argus.unit_check || argus.norm_check)
|
|
{
|
|
gesvdx_getError<STRIDED, T>(
|
|
handle, leftv, rightv, srange, m, n, dA, lda, stA, vl, vu, il, iu, dNsv, dS, stS,
|
|
dU, ldu, stU, dV, ldv, stV, difail, stF, dinfo, bc, leftvT, rightvT, mT, nT, dUT,
|
|
lduT, stUT, dVT, ldvT, stVT, hA, hNsv, hNsvRes, hS, hSres, hU, hUres, ldures, hV,
|
|
hVres, ldvres, hifail, hifailRes, hinfo, hinfoRes, &max_error, &max_errorv);
|
|
}
|
|
|
|
// collect performance data
|
|
if(argus.timing && hot_calls > 0)
|
|
{
|
|
gesvdx_getPerfData<STRIDED, T>(handle, leftv, rightv, srange, m, n, dA, lda, stA, vl,
|
|
vu, il, iu, dNsv, dS, stS, dU, ldu, stU, dV, ldv, stV,
|
|
difail, stF, dinfo, bc, hA, hNsv, hS, hU, hV, hifail,
|
|
hinfo, &gpu_time_used, &cpu_time_used, hot_calls,
|
|
argus.profile, argus.profile_kernels, argus.perf);
|
|
}
|
|
}
|
|
|
|
else
|
|
{
|
|
// memory allocations
|
|
host_strided_batch_vector<T> hA(size_A, 1, stA, bc);
|
|
device_strided_batch_vector<T> dA(size_A, 1, stA, bc);
|
|
if(size_A)
|
|
CHECK_HIP_ERROR(dA.memcheck());
|
|
|
|
// check quick return
|
|
if(n == 0 || m == 0 || bc == 0)
|
|
{
|
|
EXPECT_ROCBLAS_STATUS(rocsolver_gesvdx(STRIDED, handle, leftv, rightv, srange, m, n,
|
|
dA.data(), lda, stA, vl, vu, il, iu, dNsv.data(),
|
|
dS.data(), stS, dU.data(), ldu, stU, dV.data(),
|
|
ldv, stV, difail.data(), stF, dinfo.data(), bc),
|
|
rocblas_status_success);
|
|
if(argus.timing)
|
|
rocsolver_bench_inform(inform_quick_return);
|
|
|
|
return;
|
|
}
|
|
|
|
// check computations
|
|
if(argus.unit_check || argus.norm_check)
|
|
{
|
|
gesvdx_getError<STRIDED, T>(
|
|
handle, leftv, rightv, srange, m, n, dA, lda, stA, vl, vu, il, iu, dNsv, dS, stS,
|
|
dU, ldu, stU, dV, ldv, stV, difail, stF, dinfo, bc, leftvT, rightvT, mT, nT, dUT,
|
|
lduT, stUT, dVT, ldvT, stVT, hA, hNsv, hNsvRes, hS, hSres, hU, hUres, ldures, hV,
|
|
hVres, ldvres, hifail, hifailRes, hinfo, hinfoRes, &max_error, &max_errorv);
|
|
}
|
|
|
|
// collect performance data
|
|
if(argus.timing && hot_calls > 0)
|
|
{
|
|
gesvdx_getPerfData<STRIDED, T>(handle, leftv, rightv, srange, m, n, dA, lda, stA, vl,
|
|
vu, il, iu, dNsv, dS, stS, dU, ldu, stU, dV, ldv, stV,
|
|
difail, stF, dinfo, bc, hA, hNsv, hS, hU, hV, hifail,
|
|
hinfo, &gpu_time_used, &cpu_time_used, hot_calls,
|
|
argus.profile, argus.profile_kernels, argus.perf);
|
|
}
|
|
}
|
|
|
|
// validate results for rocsolver-test
|
|
// using 2 * min(m, n) * machine_precision as tolerance
|
|
if(argus.unit_check)
|
|
{
|
|
ROCSOLVER_TEST_CHECK(T, max_error, 2 * std::min(m, n));
|
|
if(svects)
|
|
ROCSOLVER_TEST_CHECK(T, max_errorv, 4 * std::min(m, n));
|
|
}
|
|
|
|
// output results for rocsolver-bench
|
|
if(argus.timing)
|
|
{
|
|
if(svects)
|
|
max_error = (max_error >= max_errorv) ? max_error : max_errorv;
|
|
|
|
if(!argus.perf)
|
|
{
|
|
rocsolver_bench_header("Arguments:");
|
|
if(BATCHED)
|
|
{
|
|
rocsolver_bench_output("left_svect", "right_svect", "srange", "m", "n", "lda", "vl",
|
|
"vu", "il", "iu", "strideS", "ldu", "strideU", "ldv",
|
|
"strideV", "strideF", "batch_c");
|
|
rocsolver_bench_output(leftvC, rightvC, srangeC, m, n, lda, vl, vu, il, iu, stS,
|
|
ldu, stU, ldv, stV, stF, bc);
|
|
}
|
|
else if(STRIDED)
|
|
{
|
|
rocsolver_bench_output("left_svect", "right_svect", "srange", "m", "n", "lda",
|
|
"strideA", "vl", "vu", "il", "iu", "strideS", "ldu",
|
|
"strideU", "ldv", "strideV", "strideF", "batch_c");
|
|
rocsolver_bench_output(leftvC, rightvC, srangeC, m, n, lda, stA, vl, vu, il, iu,
|
|
stS, ldu, stU, ldv, stV, stF, bc);
|
|
}
|
|
else
|
|
{
|
|
rocsolver_bench_output("left_svect", "right_svect", "srange", "m", "n", "lda", "vl",
|
|
"vu", "il", "iu", "ldu", "ldv");
|
|
rocsolver_bench_output(leftvC, rightvC, srangeC, m, n, lda, vl, vu, il, iu, ldu, ldv);
|
|
}
|
|
rocsolver_bench_header("Results:");
|
|
if(argus.norm_check)
|
|
{
|
|
rocsolver_bench_output("cpu_time_us", "gpu_time_us", "error");
|
|
rocsolver_bench_output(cpu_time_used, gpu_time_used, max_error);
|
|
}
|
|
else
|
|
{
|
|
rocsolver_bench_output("cpu_time_us", "gpu_time_us");
|
|
rocsolver_bench_output(cpu_time_used, gpu_time_used);
|
|
}
|
|
rocsolver_bench_endl();
|
|
}
|
|
else
|
|
{
|
|
if(argus.norm_check)
|
|
rocsolver_bench_output(gpu_time_used, max_error);
|
|
else
|
|
rocsolver_bench_output(gpu_time_used);
|
|
}
|
|
}
|
|
|
|
// ensure all arguments were consumed
|
|
argus.validate_consumed();
|
|
}
|
|
|
|
#define EXTERN_TESTING_GESVDX(...) extern template void testing_gesvdx<__VA_ARGS__>(Arguments&);
|
|
|
|
INSTANTIATE(EXTERN_TESTING_GESVDX, FOREACH_MATRIX_DATA_LAYOUT, FOREACH_SCALAR_TYPE, APPLY_STAMP)
|