Skip to content

[rocBLAS] Users/torrezuk/swdev 568158 syrk ex tolerance fix (#2851) #2873

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
idass1990 merged 1 commit into from
Nov 27, 2025
Merged

[rocBLAS] Users/torrezuk/swdev 568158 syrk ex tolerance fix (#2851) #2873

Show file tree
Hide file tree
Changes from all commits
Commits
Show all changes
1 commit
Select commit
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
58 changes: 42 additions & 16 deletions projects/rocblas/clients/common/cblas_interface.cpp
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -2242,26 +2242,52 @@ void ref_syrk_ex(rocblas_fill uplo,
U* C,
int64_t ldc)
{
float alpha_float = alpha;
float beta_float = beta;
if constexpr(!std::is_same_v<Tc, double>)
{
float alpha_float = alpha;
float beta_float = beta;

host_vector<float> A_float, C_float;
host_vector<float> A_float, C_float;

cast_to_buffer(transA, n, k, lda, A, A_float);
cast_to_buffer(rocblas_operation_none, n, n, ldc, C, C_float);
cast_to_buffer(transA, n, k, lda, A, A_float);
cast_to_buffer(rocblas_operation_none, n, n, ldc, C, C_float);

ref_syrk(uplo,
transA,
n,
k,
alpha_float,
(const float*)A_float.data(),
lda,
beta_float,
C_float.data(),
ldc);
ref_syrk(uplo,
transA,
n,
k,
alpha_float,
(const float*)A_float.data(),
lda,
beta_float,
C_float.data(),
ldc);

cast_from_buffer(n, n, ldc, C_float, C);
}
else
{
double alpha_double = alpha;
double beta_double = beta;

host_vector<double> A_double, C_double;

cast_to_buffer(transA, n, k, lda, A, A_double);
cast_to_buffer(rocblas_operation_none, n, n, ldc, C, C_double);

cast_from_buffer(n, n, ldc, C_float, C);
ref_syrk(uplo,
transA,
n,
k,
alpha_double,
(const double*)A_double.data(),
lda,
beta_double,
C_double.data(),
ldc);

cast_from_buffer(n, n, ldc, C_double, C);
}
}

#define INSTANTIATE_SYRK_EX_TEMPLATE(T_, U_, Tc_) \
Expand Down
4 changes: 2 additions & 2 deletions projects/rocblas/clients/include/blas_ex/testing_syrk_ex.hpp
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -372,8 +372,8 @@ void testing_syrk_ex(const Arguments& arg)
// reference is computed on floats
double tol = rocblas_handle(handle)->getArchMajor() == 11
? sum_error_tolerance_for_gfx11<Tex, Ti, To>
: sum_error_tolerance<Ti>;
tol *= K * 4;
: 4 * sum_error_tolerance<Ti>;
tol = tol * K + 2 * sum_error_tolerance<To>; // add To conversion rounding error
near_check_general<To, To_hpa>(N, N, ldc, hC_gold, hC, tol);
}
else
Expand Down
6 changes: 6 additions & 0 deletions projects/rocblas/clients/include/near.hpp
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -86,6 +86,12 @@ template <>
inline constexpr double
sum_error_tolerance_for_gfx11<rocblas_half, rocblas_half, rocblas_half> = 1 / 100.0;

template <> // syrk_ex use
inline constexpr double sum_error_tolerance_for_gfx11<double, float, float> = get_epsilon<float>();

template <> // syrk_ex use
inline constexpr double sum_error_tolerance_for_gfx11<double, float, double> = get_epsilon<float>();

template <>
inline constexpr double sum_error_tolerance_for_gfx11<rocblas_float_complex,
rocblas_float_complex,
Expand Down