Skip to content

[rocSOLVER] Update rocsolver tests to measure median time #2455

New issue

Have a question about this project? Sign up for a free GitHub account to open an issue and contact its maintainers and the community.

Sign up for GitHub

By clicking “Sign up for GitHub”, you agree to our terms of service and privacy statement. We’ll occasionally send you account related emails.

Already on GitHub? Sign in to your account

Merged
AGonzales-amd merged 9 commits into from
Nov 13, 2025
Merged

[rocSOLVER] Update rocsolver tests to measure median time #2455

Show file tree
Hide file tree
Changes from all commits
Commits
Show all changes
9 commits
Select commit Hold shift + click to select a range
8ec1c9f
rocsolver timer class
AGonzales-amd Nov 3, 2025
48ba62e
formatting
AGonzales-amd Nov 3, 2025
05a5660
update test files
AGonzales-amd Nov 4, 2025
4a5fd35
add hot_calls bound + better median
AGonzales-amd Nov 4, 2025
397ce22
update missed files
AGonzales-amd Nov 4, 2025
11cef00
change avg type name
AGonzales-amd Nov 7, 2025
7aa81e7
add stl includes
AGonzales-amd Nov 12, 2025
6866955
formatting
AGonzales-amd Nov 12, 2025
dfa8df1
formatting
AGonzales-amd Nov 12, 2025
File filter

Filter by extension

Filter by extension
Conversations
Failed to load comments.
Loading
Jump to
Jump to file
Failed to load files.
Loading
Diff view
Diff view
11 changes: 6 additions & 5 deletions projects/rocsolver/clients/common/auxiliary/testing_bdsqr.hpp
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -34,6 +34,7 @@
#include "common/misc/rocsolver.hpp"
#include "common/misc/rocsolver_arguments.hpp"
#include "common/misc/rocsolver_test.hpp"
#include "common/misc/rocsolver_timer.hpp"

template <typename T, typename S>
void bdsqr_checkBadArgs(const rocblas_handle handle,
Expand Down Expand Up @@ -434,7 +435,7 @@ void bdsqr_getPerfData(const rocblas_handle handle,
// gpu-lapack performance
hipStream_t stream;
CHECK_ROCBLAS_ERROR(rocblas_get_stream(handle, &stream));
double start;
rocsolver_timer timer;

if(profile > 0)
{
Expand All @@ -451,12 +452,12 @@ void bdsqr_getPerfData(const rocblas_handle handle,
bdsqr_initData<false, true, T>(handle, uplo, n, nv, nu, nc, dD, dE, dV, ldv, dU, ldu, dC,
ldc, dInfo, hD, hE, hV, hU, hC, hInfo, D, E, false);

start = get_time_us_sync(stream);
timer.start(stream);
rocsolver_bdsqr(handle, uplo, n, nv, nu, nc, dD.data(), dE.data(), dV.data(), ldv,
dU.data(), ldu, dC.data(), ldc, dInfo.data());
*gpu_time_used += get_time_us_sync(stream) - start;
timer.end(stream);
}
*gpu_time_used /= hot_calls;
*gpu_time_used = timer.get_combined();
}

template <typename T>
Expand Down Expand Up @@ -627,7 +628,7 @@ void testing_bdsqr(Arguments& argus)
}

// collect performance data
if(argus.timing)
if(argus.timing && hot_calls > 0)
{
host_strided_batch_vector<T> hV(size_V, 1, size_V, 1);
host_strided_batch_vector<T> hU(size_U, 1, size_U, 1);
Expand Down
11 changes: 6 additions & 5 deletions projects/rocsolver/clients/common/auxiliary/testing_bdsvdx.hpp
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -34,6 +34,7 @@
#include "common/misc/rocsolver.hpp"
#include "common/misc/rocsolver_arguments.hpp"
#include "common/misc/rocsolver_test.hpp"
#include "common/misc/rocsolver_timer.hpp"

template <typename T, typename U>
void bdsvdx_checkBadArgs(const rocblas_handle handle,
Expand Down Expand Up @@ -392,7 +393,7 @@ void bdsvdx_getPerfData(const rocblas_handle handle,
// gpu-lapack performance
hipStream_t stream;
CHECK_ROCBLAS_ERROR(rocblas_get_stream(handle, &stream));
double start;
rocsolver_timer timer;

if(profile > 0)
{
Expand All @@ -408,12 +409,12 @@ void bdsvdx_getPerfData(const rocblas_handle handle,
{
bdsvdx_initData<false, true, T>(handle, n, dD, dE, hD, hE);

start = get_time_us_sync(stream);
timer.start(stream);
rocsolver_bdsvdx(handle, uplo, svect, srange, n, dD.data(), dE.data(), vl, vu, il, iu,
dNsv.data(), dS.data(), dZ.data(), ldz, dIfail.data(), dInfo.data());
*gpu_time_used += get_time_us_sync(stream) - start;
timer.end(stream);
}
*gpu_time_used /= hot_calls;
*gpu_time_used = timer.get_combined();
}

template <typename T>
Expand Down Expand Up @@ -559,7 +560,7 @@ void testing_bdsvdx(Arguments& argus)
hIfailRes, hInfo, hInfoRes, &max_error);

// collect performance data
if(argus.timing)
if(argus.timing && hot_calls > 0)
bdsvdx_getPerfData<T>(handle, uplo, svect, srange, n, dD, dE, vl, vu, il, iu, dNsv, dS, dZ,
ldz, dIfail, dInfo, hD, hE, hNsv, hS, hZ, hInfo, &gpu_time_used,
&cpu_time_used, hot_calls, argus.profile, argus.profile_kernels,
Expand Down
11 changes: 6 additions & 5 deletions projects/rocsolver/clients/common/auxiliary/testing_labrd.hpp
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -34,6 +34,7 @@
#include "common/misc/rocsolver.hpp"
#include "common/misc/rocsolver_arguments.hpp"
#include "common/misc/rocsolver_test.hpp"
#include "common/misc/rocsolver_timer.hpp"

template <typename T, typename S, typename U>
void labrd_checkBadArgs(const rocblas_handle handle,
Expand Down Expand Up @@ -291,7 +292,7 @@ void labrd_getPerfData(const rocblas_handle handle,
// gpu-lapack performance
hipStream_t stream;
CHECK_ROCBLAS_ERROR(rocblas_get_stream(handle, &stream));
double start;
rocsolver_timer timer;

if(profile > 0)
{
Expand All @@ -308,12 +309,12 @@ void labrd_getPerfData(const rocblas_handle handle,
labrd_initData<false, true, T>(handle, m, n, nb, dA, lda, dD, dE, dTauq, dTaup, dX, ldx, dY,
ldy, hA, hD, hE, hTauq, hTaup, hX, hY);

start = get_time_us_sync(stream);
timer.start(stream);
rocsolver_labrd(handle, m, n, nb, dA.data(), lda, dD.data(), dE.data(), dTauq.data(),
dTaup.data(), dX.data(), ldx, dY.data(), ldy);
*gpu_time_used += get_time_us_sync(stream) - start;
timer.end(stream);
}
*gpu_time_used /= hot_calls;
*gpu_time_used = timer.get_combined();
}

template <typename T>
Expand Down Expand Up @@ -432,7 +433,7 @@ void testing_labrd(Arguments& argus)
hARes, hD, hE, hTauq, hTaup, hX, hXRes, hY, hYRes, &max_error);

// collect performance data
if(argus.timing)
if(argus.timing && hot_calls > 0)
labrd_getPerfData<T>(handle, m, n, nb, dA, lda, dD, dE, dTauq, dTaup, dX, ldx, dY, ldy, hA,
hD, hE, hTauq, hTaup, hX, hY, &gpu_time_used, &cpu_time_used,
hot_calls, argus.profile, argus.profile_kernels, argus.perf);
Expand Down
11 changes: 6 additions & 5 deletions projects/rocsolver/clients/common/auxiliary/testing_lacgv.hpp
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -34,6 +34,7 @@
#include "common/misc/rocsolver.hpp"
#include "common/misc/rocsolver_arguments.hpp"
#include "common/misc/rocsolver_test.hpp"
#include "common/misc/rocsolver_timer.hpp"

template <typename T, typename I>
void lacgv_checkBadArgs(const rocblas_handle handle, const I n, T dA, const I inc)
Expand Down Expand Up @@ -149,7 +150,7 @@ void lacgv_getPerfData(const rocblas_handle handle,
// gpu-lapack performance
hipStream_t stream;
CHECK_ROCBLAS_ERROR(rocblas_get_stream(handle, &stream));
double start;
rocsolver_timer timer;

if(profile > 0)
{
Expand All @@ -165,11 +166,11 @@ void lacgv_getPerfData(const rocblas_handle handle,
{
lacgv_initData<false, true, T>(handle, n, dA, inc, hA);

start = get_time_us_sync(stream);
timer.start(stream);
rocsolver_lacgv(handle, n, dA.data(), inc);
*gpu_time_used += get_time_us_sync(stream) - start;
timer.end(stream);
}
*gpu_time_used /= hot_calls;
*gpu_time_used = timer.get_combined();
}

template <typename T, typename I>
Expand Down Expand Up @@ -240,7 +241,7 @@ void testing_lacgv(Arguments& argus)
lacgv_getError<T>(handle, n, dA, inc, hA, hAr, &max_error);

// collect performance data
if(argus.timing)
if(argus.timing && hot_calls > 0)
lacgv_getPerfData<T>(handle, n, dA, inc, hA, &gpu_time_used, &cpu_time_used, hot_calls,
argus.profile, argus.profile_kernels, argus.perf);

Expand Down
11 changes: 6 additions & 5 deletions projects/rocsolver/clients/common/auxiliary/testing_larf.hpp
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -34,6 +34,7 @@
#include "common/misc/rocsolver.hpp"
#include "common/misc/rocsolver_arguments.hpp"
#include "common/misc/rocsolver_test.hpp"
#include "common/misc/rocsolver_timer.hpp"

template <typename T, typename I>
void larf_checkBadArgs(const rocblas_handle handle,
Expand Down Expand Up @@ -223,7 +224,7 @@ void larf_getPerfData(const rocblas_handle handle,
// gpu-lapack performance
hipStream_t stream;
CHECK_ROCBLAS_ERROR(rocblas_get_stream(handle, &stream));
double start;
rocsolver_timer timer;

if(profile > 0)
{
Expand All @@ -239,11 +240,11 @@ void larf_getPerfData(const rocblas_handle handle,
{
larf_initData<false, true, T>(handle, side, m, n, dx, inc, dt, dA, lda, xx, hx, ht, hA);

start = get_time_us_sync(stream);
timer.start(stream);
rocsolver_larf(handle, side, m, n, dx.data(), inc, dt.data(), dA.data(), lda);
*gpu_time_used += get_time_us_sync(stream) - start;
timer.end(stream);
}
*gpu_time_used /= hot_calls;
*gpu_time_used = timer.get_combined();
}

template <typename T, typename I>
Expand Down Expand Up @@ -343,7 +344,7 @@ void testing_larf(Arguments& argus)
larf_getError<T>(handle, side, m, n, dx, inc, dt, dA, lda, xx, hx, ht, hA, hAr, &max_error);

// collect performance data
if(argus.timing)
if(argus.timing && hot_calls > 0)
larf_getPerfData<T>(handle, side, m, n, dx, inc, dt, dA, lda, xx, hx, ht, hA,
&gpu_time_used, &cpu_time_used, hot_calls, argus.profile,
argus.profile_kernels, argus.perf);
Expand Down
11 changes: 6 additions & 5 deletions projects/rocsolver/clients/common/auxiliary/testing_larfb.hpp
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -34,6 +34,7 @@
#include "common/misc/rocsolver.hpp"
#include "common/misc/rocsolver_arguments.hpp"
#include "common/misc/rocsolver_test.hpp"
#include "common/misc/rocsolver_timer.hpp"

template <typename T>
void larfb_checkBadArgs(const rocblas_handle handle,
Expand Down Expand Up @@ -351,7 +352,7 @@ void larfb_getPerfData(const rocblas_handle handle,
// gpu-lapack performance
hipStream_t stream;
CHECK_ROCBLAS_ERROR(rocblas_get_stream(handle, &stream));
double start;
rocsolver_timer timer;

if(profile > 0)
{
Expand All @@ -368,12 +369,12 @@ void larfb_getPerfData(const rocblas_handle handle,
larfb_initData<false, true, T>(handle, side, trans, direct, storev, m, n, k, dV, ldv, dT,
ldt, dA, lda, hV, hT, hA, hW, sizeW);

start = get_time_us_sync(stream);
timer.start(stream);
rocsolver_larfb(handle, side, trans, direct, storev, m, n, k, dV.data(), ldv, dT.data(),
ldt, dA.data(), lda);
*gpu_time_used += get_time_us_sync(stream) - start;
timer.end(stream);
}
*gpu_time_used /= hot_calls;
*gpu_time_used = timer.get_combined();
}

template <typename T>
Expand Down Expand Up @@ -488,7 +489,7 @@ void testing_larfb(Arguments& argus)
hV, hT, hA, hAr, &max_error);

// collect performance data
if(argus.timing)
if(argus.timing && hot_calls > 0)
larfb_getPerfData<T>(handle, side, trans, direct, storev, m, n, k, dV, ldv, dT, ldt, dA,
lda, hV, hT, hA, &gpu_time_used, &cpu_time_used, hot_calls,
argus.profile, argus.profile_kernels, argus.perf);
Expand Down
11 changes: 6 additions & 5 deletions projects/rocsolver/clients/common/auxiliary/testing_larfg.hpp
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -34,6 +34,7 @@
#include "common/misc/rocsolver.hpp"
#include "common/misc/rocsolver_arguments.hpp"
#include "common/misc/rocsolver_test.hpp"
#include "common/misc/rocsolver_timer.hpp"

template <typename T, typename I>
void larfg_checkBadArgs(const rocblas_handle handle, const I n, T da, T dx, const I inc, T dt)
Expand Down Expand Up @@ -174,7 +175,7 @@ void larfg_getPerfData(const rocblas_handle handle,
// gpu-lapack performance
hipStream_t stream;
CHECK_ROCBLAS_ERROR(rocblas_get_stream(handle, &stream));
double start;
rocsolver_timer timer;

if(profile > 0)
{
Expand All @@ -190,11 +191,11 @@ void larfg_getPerfData(const rocblas_handle handle,
{
larfg_initData<false, true, T>(handle, n, da, dx, inc, dt, ha, hx, ht);

start = get_time_us_sync(stream);
timer.start(stream);
rocsolver_larfg(handle, n, da.data(), dx.data(), inc, dt.data());
*gpu_time_used += get_time_us_sync(stream) - start;
timer.end(stream);
}
*gpu_time_used /= hot_calls;
*gpu_time_used = timer.get_combined();
}

template <typename T, typename I>
Expand Down Expand Up @@ -277,7 +278,7 @@ void testing_larfg(Arguments& argus)
larfg_getError<T>(handle, n, da, dx, inc, dt, ha, hx, hxr, ht, &max_error);

// collect performance data
if(argus.timing)
if(argus.timing && hot_calls > 0)
larfg_getPerfData<T>(handle, n, da, dx, inc, dt, ha, hx, ht, &gpu_time_used, &cpu_time_used,
hot_calls, argus.profile, argus.profile_kernels, argus.perf);

Expand Down
11 changes: 6 additions & 5 deletions projects/rocsolver/clients/common/auxiliary/testing_larft.hpp
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -34,6 +34,7 @@
#include "common/misc/rocsolver.hpp"
#include "common/misc/rocsolver_arguments.hpp"
#include "common/misc/rocsolver_test.hpp"
#include "common/misc/rocsolver_timer.hpp"

template <typename T>
void larft_checkBadArgs(const rocblas_handle handle,
Expand Down Expand Up @@ -258,7 +259,7 @@ void larft_getPerfData(const rocblas_handle handle,
// gpu-lapack performance
hipStream_t stream;
CHECK_ROCBLAS_ERROR(rocblas_get_stream(handle, &stream));
double start;
rocsolver_timer timer;

if(profile > 0)
{
Expand All @@ -275,11 +276,11 @@ void larft_getPerfData(const rocblas_handle handle,
larft_initData<false, true, T>(handle, direct, storev, n, k, dV, ldv, dt, dT, ldt, hV, ht,
hT, hw, size_w);

start = get_time_us_sync(stream);
timer.start(stream);
rocsolver_larft(handle, direct, storev, n, k, dV.data(), ldv, dt.data(), dT.data(), ldt);
*gpu_time_used += get_time_us_sync(stream) - start;
timer.end(stream);
}
*gpu_time_used /= hot_calls;
*gpu_time_used = timer.get_combined();
}

template <typename T>
Expand Down Expand Up @@ -372,7 +373,7 @@ void testing_larft(Arguments& argus)
&max_error);

// collect performance data
if(argus.timing)
if(argus.timing && hot_calls > 0)
larft_getPerfData<T>(handle, direct, storev, n, k, dV, ldv, dt, dT, ldt, hV, ht, hT,
&gpu_time_used, &cpu_time_used, hot_calls, argus.profile,
argus.profile_kernels, argus.perf);
Expand Down
9 changes: 5 additions & 4 deletions projects/rocsolver/clients/common/auxiliary/testing_lasr.hpp
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -34,6 +34,7 @@
#include "common/misc/rocsolver.hpp"
#include "common/misc/rocsolver_arguments.hpp"
#include "common/misc/rocsolver_test.hpp"
#include "common/misc/rocsolver_timer.hpp"

template <typename T, typename S>
void lasr_checkBadArgs(const rocblas_handle handle,
Expand Down Expand Up @@ -240,7 +241,7 @@ void lasr_getPerfData(const rocblas_handle handle,
// gpu-lapack performance
hipStream_t stream;
CHECK_ROCBLAS_ERROR(rocblas_get_stream(handle, &stream));
double start;
rocsolver_timer timer;

if(profile > 0)
{
Expand All @@ -256,11 +257,11 @@ void lasr_getPerfData(const rocblas_handle handle,
{
lasr_initData<false, true, T>(handle, side, pivot, direct, m, n, dC, dS, dA, lda, hC, hS, hA);

start = get_time_us_sync(stream);
timer.start(stream);
rocsolver_lasr(handle, side, pivot, direct, m, n, dC.data(), dS.data(), dA.data(), lda);
*gpu_time_used += get_time_us_sync(stream) - start;
timer.end(stream);
}
*gpu_time_used /= hot_calls;
*gpu_time_used = timer.get_combined();
}

template <typename T>
Expand Down
Loading