Skip to content

RISC-V: fix zero-output-stride AXPY accumulation #6195

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

Closed
SmaCow666 wants to merge 2 commits into from
Closed

RISC-V: fix zero-output-stride AXPY accumulation #6195

Show file tree
Hide file tree
Changes from all commits
Commits
Show all changes
2 commits
Select commit Hold shift + click to select a range
a05fd6e
RISC-V: fix zero-output-stride AXPY accumulation
Oct 29, 2026
816d78b
Merge pull request #1 from OpenMathLib/develop
SmaCow666 Oct 29, 2026
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
13 changes: 13 additions & 0 deletions kernel/riscv64/axpy_vector.c
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -161,6 +161,17 @@ int CNAME(BLASLONG n, BLASLONG dummy0, BLASLONG dummy1, FLOAT da, FLOAT *x, BLAS
j += gvl;
}
} else if( 0 == inc_y ) {
#if defined(RISCV64_ZVL256B)
/* Zero output stride is an ordered update of one scalar. Do not
broadcast its initial value into every reduction lane. */
for (BLASLONG k = 0; k < n; ++k) {
#if defined(DOUBLE)
y[0] = __builtin_fma(da, x[k * inc_x], y[0]);
#else
y[0] = __builtin_fmaf(da, x[k * inc_x], y[0]);
#endif
}
#else
BLASLONG stride_x = inc_x * sizeof(FLOAT);
size_t in_vl = VSETVL(n);
vy0 = VFMVVF_FLOAT( y[0], in_vl );
Expand All @@ -173,6 +184,8 @@ int CNAME(BLASLONG n, BLASLONG dummy0, BLASLONG dummy1, FLOAT da, FLOAT *x, BLAS
FLOAT_V_M1_T v_res = VFMVVF_FLOAT_M1( 0.0f, 1 );
v_res = VFREDSUMVS_FLOAT( vy0, v_res, in_vl );
y[0] = EXTRACT_FLOAT(v_res);
#endif

}else{
stride_x = inc_x * sizeof(FLOAT);
stride_y = inc_y * sizeof(FLOAT);
Expand Down
55 changes: 55 additions & 0 deletions utest/test_axpy.c
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -157,3 +157,58 @@ CTEST(axpy,caxpy_incx_0)
}
#endif



#ifdef BUILD_DOUBLE
CTEST(axpy,daxpy_zero_output_stride)
{
const blasint lengths[] = {1, 2, 9, 17, 33};
const blasint strides[] = {1, 2, -1, -2};
const double seeds[] = {-2, 0, 2};
const double alphas[] = {0, 1, -1, 0.5};
int ni, si, yi, ai, k;
for (ni = 0; ni < 5; ++ni)
for (si = 0; si < 4; ++si)
for (yi = 0; yi < 3; ++yi)
for (ai = 0; ai < 4; ++ai) {
blasint n = lengths[ni], incx = strides[si], incy = 0;
double alpha = alphas[ai], x[65], y[3] = {99, 0, 99};
double expected = seeds[yi] + n * alpha;
y[1] = seeds[yi];
for (k = 0; k < 65; ++k) x[k] = 1;
BLASFUNC(daxpy)(&n, &alpha, x, &incx, y + 1, &incy);
ASSERT_DBL_NEAR_TOL(expected, y[1], DOUBLE_EPS);
ASSERT_DBL_NEAR_TOL(99, y[0], DOUBLE_EPS);
ASSERT_DBL_NEAR_TOL(99, y[2], DOUBLE_EPS);
for (k = 0; k < 65; ++k)
ASSERT_DBL_NEAR_TOL(1, x[k], DOUBLE_EPS);
}
}
#endif

#ifdef BUILD_SINGLE
CTEST(axpy,saxpy_zero_output_stride)
{
const blasint lengths[] = {1, 2, 9, 17, 33};
const blasint strides[] = {1, 2, -1, -2};
const float seeds[] = {-2, 0, 2};
const float alphas[] = {0, 1, -1, 0.5};
int ni, si, yi, ai, k;
for (ni = 0; ni < 5; ++ni)
for (si = 0; si < 4; ++si)
for (yi = 0; yi < 3; ++yi)
for (ai = 0; ai < 4; ++ai) {
blasint n = lengths[ni], incx = strides[si], incy = 0;
float alpha = alphas[ai], x[65], y[3] = {99, 0, 99};
float expected = seeds[yi] + n * alpha;
y[1] = seeds[yi];
for (k = 0; k < 65; ++k) x[k] = 1;
BLASFUNC(saxpy)(&n, &alpha, x, &incx, y + 1, &incy);
ASSERT_DBL_NEAR_TOL(expected, y[1], DOUBLE_EPS);
ASSERT_DBL_NEAR_TOL(99, y[0], DOUBLE_EPS);
ASSERT_DBL_NEAR_TOL(99, y[2], DOUBLE_EPS);
for (k = 0; k < 65; ++k)
ASSERT_DBL_NEAR_TOL(1, x[k], DOUBLE_EPS);
}
}
#endif