Skip to content

tweak to functor implementing gemm #1810

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
oleksandr-pavlyk merged 1 commit into from
Aug 23, 2024
Merged

tweak to functor implementing gemm #1810

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
22 changes: 18 additions & 4 deletions dpctl/tensor/libtensor/include/kernels/linalg_functions/gemm.hpp
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -970,6 +970,7 @@ class GemmBatchFunctorThreadNM_vecm
size_t i = block_i * wg_delta_n * wi_delta_n;
size_t j = block_j * wg_delta_m * wi_total_delta_m;

using slmA_t = typename LocAccT1::value_type;
using slmB_t = typename LocAccT2::value_type;

const size_t a_st0 = k;
Expand Down Expand Up @@ -1057,16 +1058,29 @@ class GemmBatchFunctorThreadNM_vecm
const std::uint32_t lo_lhs_st_k = (wg_delta_n * wi_delta_n);
const std::uint32_t lo_rhs_rk_k = (wg_delta_m * wi_delta_m_vecs);
for (std::uint32_t pr_k = 0; pr_k < wi_delta_k; ++pr_k) {
std::array<slmA_t, wi_delta_n> pr_lhs{};
#pragma unroll
for (std::uint32_t pr_i = 0; pr_i < wi_delta_n; ++pr_i) {
pr_lhs[pr_i] =
local_lhs_block[pr_k * lo_lhs_st_k +
(local_i + pr_i * wg_delta_n)];
}

std::array<slmB_t, wi_delta_m_vecs> pr_rhs{};
#pragma unroll
for (std::uint32_t pr_j = 0; pr_j < wi_delta_m_vecs; ++pr_j) {
pr_rhs[pr_j] =
local_rhs_block[pr_k * lo_rhs_rk_k +
(local_j + pr_j * wg_delta_m)];
}

#pragma unroll
for (std::uint32_t pr_i = 0; pr_i < wi_delta_n; ++pr_i) {
#pragma unroll
for (std::uint32_t pr_j = 0; pr_j < wi_delta_m_vecs; ++pr_j)
{
private_C[pr_i * wi_delta_m_vecs + pr_j] +=
local_lhs_block[pr_k * lo_lhs_st_k +
(local_i + pr_i * wg_delta_n)] *
local_rhs_block[pr_k * lo_rhs_rk_k +
(local_j + pr_j * wg_delta_m)];
pr_lhs[pr_i] * pr_rhs[pr_j];
}
}
}
Expand Down