From b4e3e1c371d8091e86ee1c6e704057559bbe1554 Mon Sep 17 00:00:00 2001 From: Ioan-Cristian Szabo Date: Thu, 30 Nov 2017 17:17:17 +0000 Subject: COMPMID-617: Add validate support for NEON FullyConnectedLayer Change-Id: I08987022c8d4cc335c00b8af27bd3edb8fe64d3b Reviewed-on: https://eu-gerrit-1.euhpc.arm.com/111596 Tested-by: Jenkins Reviewed-by: Alexander Gilday Reviewed-by: Anthony Barbier --- src/runtime/NEON/functions/NEGEMM.cpp | 8 ++++++-- 1 file changed, 6 insertions(+), 2 deletions(-) (limited to 'src/runtime/NEON/functions/NEGEMM.cpp') diff --git a/src/runtime/NEON/functions/NEGEMM.cpp b/src/runtime/NEON/functions/NEGEMM.cpp index 48a0d2af1c..05907bab07 100644 --- a/src/runtime/NEON/functions/NEGEMM.cpp +++ b/src/runtime/NEON/functions/NEGEMM.cpp @@ -120,7 +120,7 @@ void NEGEMM::configure(const ITensor *a, const ITensor *b, const ITensor *c, ITe #endif /* defined(__aarch64__) */ { // Configure the matrix multiply kernel - _mm_kernel.configure(a, b, d, alpha); + _mm_kernel.configure(a, b, d, alpha, false); } // Configure matrix addition kernel @@ -212,6 +212,10 @@ void NEGEMM::configure(const ITensor *a, const ITensor *b, const ITensor *c, ITe _memory_group.manage(&_tmp_a); _memory_group.manage(&_tmp_b); + int m = a->info()->dimension(1); + int n = b->info()->dimension(0); + int k = a->info()->dimension(0); + // Configure interleave kernel _interleave_kernel.configure(a, &_tmp_a); @@ -219,7 +223,7 @@ void NEGEMM::configure(const ITensor *a, const ITensor *b, const ITensor *c, ITe _transpose_kernel.configure(b, &_tmp_b); // Configure matrix multiplication kernel - _mm_kernel.configure(&_tmp_a, &_tmp_b, d, alpha); + _mm_kernel.configure(&_tmp_a, &_tmp_b, d, alpha, true, GEMMReshapeInfo(m, n, k)); // Allocate once the all configure methods have been called _tmp_a.allocator()->allocate(); -- cgit v1.2.1