From b5311a63d9bc1d8c9265aba5a9e031726cc512b3 Mon Sep 17 00:00:00 2001 From: Gian Marco Date: Wed, 13 Dec 2017 12:48:03 +0000 Subject: COMPMID-556 - Enabled optimized path for Bifrost in CLGEMM - Enabled optimized path for Bifrost when the input matrices have not been reshaped Change-Id: I92c5591846ce7ea9116fecc434a0e859f3f2d579 Reviewed-on: https://eu-gerrit-1.euhpc.arm.com/113083 Reviewed-by: Georgios Pinitas Reviewed-by: Anthony Barbier Tested-by: BSG Visual Compute Jenkins server to access repositories on http://mpd-gerrit.cambridge.arm.com --- src/runtime/CL/functions/CLGEMM.cpp | 6 +++--- 1 file changed, 3 insertions(+), 3 deletions(-) (limited to 'src/runtime/CL/functions/CLGEMM.cpp') diff --git a/src/runtime/CL/functions/CLGEMM.cpp b/src/runtime/CL/functions/CLGEMM.cpp index a81d1138c0..ca0228fcdb 100644 --- a/src/runtime/CL/functions/CLGEMM.cpp +++ b/src/runtime/CL/functions/CLGEMM.cpp @@ -65,6 +65,9 @@ void CLGEMM::configure(const ICLTensor *a, const ICLTensor *b, const ICLTensor * const ICLTensor *matrix_a = a; const ICLTensor *matrix_b = b; + // Set the target for the matrix multiply kernel + _mm_kernel.set_target(CLScheduler::get().target()); + if(_is_interleaved_transposed) { matrix_a = &_tmp_a; @@ -95,9 +98,6 @@ void CLGEMM::configure(const ICLTensor *a, const ICLTensor *b, const ICLTensor * // Configure transpose kernel _transpose_kernel.configure(b, &_tmp_b); - - // Configure matrix multiply kernel - _mm_kernel.set_target(CLScheduler::get().target()); } _mm_kernel.configure(matrix_a, matrix_b, output, alpha, _is_interleaved_transposed); -- cgit v1.2.1