diff options
author | Cedric Nugteren <web@cedricnugteren.nl> | 2017-03-11 16:02:45 +0100 |
---|---|---|
committer | Cedric Nugteren <web@cedricnugteren.nl> | 2017-03-11 16:02:45 +0100 |
commit | 7b8f8fce6808f2095a68afe97256db7a78f819fa (patch) | |
tree | 98f394e504740389cebf1b9061e269645e51ec7d /src/routines/level3/xgemm.cpp | |
parent | 49e04c7fce8fed45559e143137cef3a1a36328cc (diff) |
Added initial naive version of the batched GEMM routine based on the direct GEMM kernel
Diffstat (limited to 'src/routines/level3/xgemm.cpp')
-rw-r--r-- | src/routines/level3/xgemm.cpp | 22 |
1 files changed, 11 insertions, 11 deletions
diff --git a/src/routines/level3/xgemm.cpp b/src/routines/level3/xgemm.cpp index dc8c64bc..658b22d0 100644 --- a/src/routines/level3/xgemm.cpp +++ b/src/routines/level3/xgemm.cpp @@ -104,19 +104,19 @@ void Xgemm<T>::DoGemm(const Layout layout, // Selects which version of GEMM to run const auto do_gemm_direct = (m * n * k < db_["XGEMM_MIN_INDIRECT_SIZE"]); if (do_gemm_direct) { // for small sizes (single kernel) - return GemmDirect(m, n, k, alpha, - a_buffer, a_offset, a_ld, b_buffer, b_offset, b_ld, beta, - c_buffer, c_offset, c_ld, - a_do_transpose, b_do_transpose, c_do_transpose, a_conjugate, b_conjugate); + GemmDirect(m, n, k, alpha, + a_buffer, a_offset, a_ld, b_buffer, b_offset, b_ld, beta, + c_buffer, c_offset, c_ld, + a_do_transpose, b_do_transpose, c_do_transpose, a_conjugate, b_conjugate); } else { // for larger sizes (pre/post-processing plus a very fast kernel) - return GemmIndirect(m, n, k, alpha, - a_buffer, a_offset, a_ld, b_buffer, b_offset, b_ld, beta, - c_buffer, c_offset, c_ld, - a_do_transpose, b_do_transpose, c_do_transpose, a_conjugate, b_conjugate, - a_one, a_two, a_want_rotated, - b_one, b_two, b_want_rotated, - c_one, c_two, c_want_rotated); + GemmIndirect(m, n, k, alpha, + a_buffer, a_offset, a_ld, b_buffer, b_offset, b_ld, beta, + c_buffer, c_offset, c_ld, + a_do_transpose, b_do_transpose, c_do_transpose, a_conjugate, b_conjugate, + a_one, a_two, a_want_rotated, + b_one, b_two, b_want_rotated, + c_one, c_two, c_want_rotated); } } |