diff options
author | Michele Di Giorgio <michele.digiorgio@arm.com> | 2021-07-26 13:18:50 +0100 |
---|---|---|
committer | Georgios Pinitas <georgios.pinitas@arm.com> | 2021-09-07 13:44:08 +0000 |
commit | aed63ee175e0d64c934389e9d1b2edd0cb1a5cdd (patch) | |
tree | 8f025f849e863b9cdec1d6b889bc463e6c4f78d1 /src/core/NEON/kernels/arm_gemm/gemm_interleaved.hpp | |
parent | 58d3c5a7df769def499806e4d26cea518add161a (diff) | |
download | ComputeLibrary-aed63ee175e0d64c934389e9d1b2edd0cb1a5cdd.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
Change-Id: I96e6f3868f51785c9700a3ef6a1fe7b05747862c
Signed-off-by: Michele Di Giorgio <michele.digiorgio@arm.com>
Reviewed-on: https://review.mlplatform.org/c/ml/ComputeLibrary/+/6162
Tested-by: Arm Jenkins <bsgcomp@arm.com>
Comments-Addressed: Arm Jenkins <bsgcomp@arm.com>
Reviewed-by: Georgios Pinitas <georgios.pinitas@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 | 6 |
1 files changed, 5 insertions, 1 deletions
diff --git a/src/core/NEON/kernels/arm_gemm/gemm_interleaved.hpp b/src/core/NEON/kernels/arm_gemm/gemm_interleaved.hpp index 5639cb4182..c75c320a6b 100644 --- a/src/core/NEON/kernels/arm_gemm/gemm_interleaved.hpp +++ b/src/core/NEON/kernels/arm_gemm/gemm_interleaved.hpp @@ -923,7 +923,7 @@ public: return (x_size * _Ktotal * _nmulti * sizeof(Toi)) + get_col_sum_size(); } - void pretranspose_B_array(void *in_buffer, const To *B, const int ldb, const int B_multi_stride) override { + void requantize_bias(void *in_buffer, const To *B, const int ldb, const int B_multi_stride) override { if (std::is_same<OutputStage, Requantize32>::value) { col_bias = reinterpret_cast<int32_t *>(in_buffer); @@ -934,6 +934,10 @@ public: compute_col_sums(*qp_ptr, _Nsize, _Ksize * _Ksections, B + (i * B_multi_stride), ldb, col_bias + (i * _Nsize), _Ksize * _Ksections, i, 0); } } + } + + void pretranspose_B_array(void *in_buffer, const To *B, const int ldb, const int B_multi_stride) override { + requantize_bias(in_buffer, B, ldb, B_multi_stride); // Put the transposed data after the column sums - in non-transposing cases get_col_sum_size() == 0 uintptr_t buffer_int = reinterpret_cast<uintptr_t>(in_buffer); |