commit a86433a64ab25c2ea8e274bd2f357a9709636f5b
author Michele Di Giorgio <michele.digiorgio@arm.com> Wed Jul 28 14:10:47 2021 +0100
committer Georgios Pinitas <georgios.pinitas@arm.com> Wed Jul 28 16:31:53 2021 +0000
tree 48978522014f8d0423522ec37f2d491bee110944
parent 120c87b1622fa2472d27e43e5b7d861fb9735d9c

Reduce binary footprint of CpuConvertFullyConnectedWeightsKernel

Binary size reduction for this kernel is almost 50%.

Also remove unused NEConvertFullyConnectedWeightsManaged.

Change-Id: Ia46a1342a0737397b4aac2578d963c2ebb7446e3
Signed-off-by: Michele Di Giorgio <michele.digiorgio@arm.com>
Reviewed-on: https://review.mlplatform.org/c/ml/ComputeLibrary/+/6011
Reviewed-by: Giorgio Arena <giorgio.arena@arm.com>
Comments-Addressed: Arm Jenkins <bsgcomp@arm.com>
Tested-by: Arm Jenkins <bsgcomp@arm.com>

src/core/cpu/kernels/CpuConvertFullyConnectedWeightsKernel.cpp

diff --git a/src/core/cpu/kernels/CpuConvertFullyConnectedWeightsKernel.cpp b/src/core/cpu/kernels/CpuConvertFullyConnectedWeightsKernel.cpp
index 5bf70dc..5406356 100644
--- a/src/core/cpu/kernels/CpuConvertFullyConnectedWeightsKernel.cpp
+++ b/src/core/cpu/kernels/CpuConvertFullyConnectedWeightsKernel.cpp
@@ -81,22 +81,6 @@
     return Status{};
 }
 
-template <typename T>
-void CpuConvertFullyConnectedWeightsKernel::run_convert_fc_weights(const ITensor *in, ITensor *out, const Window &window)
-{
-    const unsigned int dst_stride_x = out->info()->strides_in_bytes().x();
-    const unsigned int dst_stride_y = out->info()->strides_in_bytes().y();
-
-    Iterator input(in, window);
-    Iterator output(out, window);
-
-    execute_window_loop(window, [&](const Coordinates & id)
-    {
-        *reinterpret_cast<T *>(output.ptr() + id.x() * dst_stride_x + (id.y() % _factor1 * _factor2 + id.y() / _factor1) * dst_stride_y) = *reinterpret_cast<T *>(input.ptr());
-    },
-    input);
-}
-
 void CpuConvertFullyConnectedWeightsKernel::run_op(ITensorPack &tensors, const Window &window, const ThreadInfo &info)
 {
     ARM_COMPUTE_UNUSED(info);
@@ -106,21 +90,18 @@
     const auto src = tensors.get_const_tensor(TensorType::ACL_SRC);
     auto       dst = tensors.get_tensor(TensorType::ACL_DST);
 
-    switch(src->info()->element_size())
+    const unsigned int dst_stride_x = dst->info()->strides_in_bytes().x();
+    const unsigned int dst_stride_y = dst->info()->strides_in_bytes().y();
+    const unsigned int element_size = src->info()->element_size();
+
+    Iterator input(src, window);
+    Iterator output(dst, window);
+
+    execute_window_loop(window, [&](const Coordinates & id)
     {
-        case 1:
-            run_convert_fc_weights<uint8_t>(src, dst, window);
-            break;
-        case 2:
-            run_convert_fc_weights<uint16_t>(src, dst, window);
-            break;
-        case 4:
-            run_convert_fc_weights<uint32_t>(src, dst, window);
-            break;
-        default:
-            ARM_COMPUTE_ERROR("Data type not supported.");
-            break;
-    }
+        memcpy(output.ptr() + id.x() * dst_stride_x + (id.y() % _factor1 * _factor2 + id.y() / _factor1) * dst_stride_y, input.ptr(), element_size);
+    },
+    input);
 }
 
 const char *CpuConvertFullyConnectedWeightsKernel::name() const