diff options
| author | Georgios Pinitas <georgios.pinitas@arm.com> | 2019-06-27 17:00:52 +0100 |
|---|---|---|
| committer | Georgios Pinitas <georgios.pinitas@arm.com> | 2019-07-26 11:55:15 +0000 |
| commit | cfa2bba98169cb5ab1945462514be1b6badf7d98 (patch) | |
| tree | 1635e6e9463e9798c7195f0aa71b5df3f2650df1 /src/core/NEON/kernels/arm_gemm/gemm_interleaved.hpp | |
| parent | f59b16f42ef68bde877b70816ffb953d64c8baa3 (diff) | |
| download | ComputeLibrary-cfa2bba98169cb5ab1945462514be1b6badf7d98.tar.gz | |
COMPMID-2178: Update GEMM assembly code.
Perform offset reduction and requantization within the assembly wrapper.
Change-Id: I5d5b3e1f6f9ef4c71805362c57f88ff199c027a3
Signed-off-by: Georgios Pinitas <georgios.pinitas@arm.com>
Reviewed-on: https://review.mlplatform.org/c/1541
Comments-Addressed: Pablo Marquez <pablo.tello@arm.com>
Reviewed-by: Gian Marco Iodice <gianmarco.iodice@arm.com>
Comments-Addressed: Arm Jenkins <bsgcomp@arm.com>
Tested-by: Arm Jenkins <bsgcomp@arm.com>
Diffstat (limited to 'src/core/NEON/kernels/arm_gemm/gemm_interleaved.hpp')
| -rw-r--r-- | src/core/NEON/kernels/arm_gemm/gemm_interleaved.hpp | 10 |
1 files changed, 5 insertions, 5 deletions
diff --git a/src/core/NEON/kernels/arm_gemm/gemm_interleaved.hpp b/src/core/NEON/kernels/arm_gemm/gemm_interleaved.hpp index a7731666ec..faff9acd2e 100644 --- a/src/core/NEON/kernels/arm_gemm/gemm_interleaved.hpp +++ b/src/core/NEON/kernels/arm_gemm/gemm_interleaved.hpp @@ -318,10 +318,10 @@ public: /* Constructor */ GemmInterleaved(const GemmArgs<Tr> &args) - : _ci(args._ci), _Msize(args._Msize), _Nsize(args._Nsize), _Ksize(args._Ksize), - _nbatches(args._nbatches), _nmulti(args._nmulti), _trA(args._trA), _trB(args._trB), - _alpha(args._alpha), _beta(args._beta), _maxthreads(args._maxthreads), _nthreads(args._maxthreads), - _pretransposed(args._pretransposed_hint) { + : _ci(args._ci), _Msize(args._Msize), _Nsize(args._Nsize), _Ksize(args._Ksize), + _nbatches(args._nbatches), _nmulti(args._nmulti), _trA(args._trA), _trB(args._trB), + _alpha(args._alpha), _beta(args._beta), _maxthreads(args._maxthreads), _nthreads(args._maxthreads), + _pretransposed(args._pretransposed_hint) { const unsigned int L1_size = _ci->get_L1_cache_size(); const unsigned int L2_size = _ci->get_L2_cache_size(); @@ -356,7 +356,7 @@ public: // x_block: Work out how many rows (of length k_block) will fit in the L2 // Don't allocate more than 90% of the L2 to allow for overheads, and subtract off the L1 contents. _x_block = (((L2_size * 9) / 10) - (_k_block * sizeof(Toi) * (strategy::out_width() + strategy::out_height()))) / - (sizeof(Toi) * _k_block); + (sizeof(Toi) * _k_block); // Needs to be (at least a single) multiple of the kernel output width. _x_block /= strategy::out_width(); |