diff options
| author | Giorgio Arena <giorgio.arena@arm.com> | 2021-09-24 14:04:27 +0100 |
|---|---|---|
| committer | Giorgio Arena <giorgio.arena@arm.com> | 2021-09-29 10:31:08 +0000 |
| commit | 63e0beb9fb9646407d123e830165546e9129e95d (patch) | |
| tree | 9bfe80e8d853327a82f9f622d89c3b43df0400f4 /src/cpu/kernels | |
| parent | b1ba1e33f2b03b211f561123559c24517c0e5865 (diff) | |
| download | ComputeLibrary-63e0beb9fb9646407d123e830165546e9129e95d.tar.gz | |
Add support for non-constant weights and biases in CpuFullyConnected
Changing the approach for specifying that weights and biases tensors are
non-constant by making it a member of TensorInfo rather than an option
of the functions.
Resolves: COMPMID-4222, COMPMID-4811
Signed-off-by: Giorgio Arena <giorgio.arena@arm.com>
Change-Id: I9b0081ccbcf8271ce029ba6755563d64c59e1d32
Reviewed-on: https://review.mlplatform.org/c/ml/ComputeLibrary/+/6313
Tested-by: Arm Jenkins <bsgcomp@arm.com>
Reviewed-by: Pablo Marquez Tello <pablo.tello@arm.com>
Reviewed-by: Michele Di Giorgio <michele.digiorgio@arm.com>
Comments-Addressed: Arm Jenkins <bsgcomp@arm.com>
Diffstat (limited to 'src/cpu/kernels')
| -rw-r--r-- | src/cpu/kernels/assembly/gemm_common.hpp | 3 |
1 files changed, 3 insertions, 0 deletions
diff --git a/src/cpu/kernels/assembly/gemm_common.hpp b/src/cpu/kernels/assembly/gemm_common.hpp index 378f1041be..ece9ca5802 100644 --- a/src/cpu/kernels/assembly/gemm_common.hpp +++ b/src/cpu/kernels/assembly/gemm_common.hpp @@ -212,6 +212,9 @@ public: /*** "Pretransposed" interface ***/ + /* Compute col sums over all columns */ + virtual void requantize_bias(void *, const To *, const int, const int) {}; + /* Perform pretranspose - the void * passed in must remain allocated for the duration of any execute calls. */ /* Arguments are: output buffer pointer, source pointer, source row stride, source multi stride */ virtual void pretranspose_B_array(void *, const To *, const int, const int) {}; |