commit 36a75dafdbe6d6a3a6f50bd075fe01f5b7dace38
author Renato Arantes <renato.arantes@arm.com> Fri Jan 26 17:31:18 2024 +0000
committer Renato Barros Arantes <renato.arantes@arm.com> Thu Mar 21 11:15:30 2024 +0000
tree 0701d615ef30444b9d0789db691b59b81fd9e86e
parent d2191150736dde66d79eb97e0c8ee506eef3c8fc

[ONCPUML-1451] Add matmul kernel to enable bf16 to bf16 operations via PyTorch® autocast() function

The full range of tests must be added with [MLINFSW-482] epic due to the lack of reordering kernels implemented in Acl.

Co-Authored-By: David Mansell <David.Mansell@arm.com>
Change-Id: I820d316295a1ec94fdc89c37e4144a268f914c36
Signed-off-by: Renato Arantes <renato.arantes@arm.com>
Reviewed-on: https://review.mlplatform.org/c/ml/ComputeLibrary/+/11169
Tested-by: Arm Jenkins <bsgcomp@arm.com>
Reviewed-by: Gunes Bayir <gunes.bayir@arm.com>
Comments-Addressed: Arm Jenkins <bsgcomp@arm.com>
Benchmark: Arm Jenkins <bsgcomp@arm.com>

tests/validation/reference/ActivationLayer.cpp

diff --git a/tests/validation/reference/ActivationLayer.cpp b/tests/validation/reference/ActivationLayer.cpp
index 664b969..2172362 100644
--- a/tests/validation/reference/ActivationLayer.cpp
+++ b/tests/validation/reference/ActivationLayer.cpp
@@ -1,5 +1,5 @@
 /*
- * Copyright (c) 2017-2020 Arm Limited.
+ * Copyright (c) 2017-2020,2024 Arm Limited.
  *
  * SPDX-License-Identifier: MIT
  *
@@ -24,6 +24,7 @@
 #include "ActivationLayer.h"
 
 #include "arm_compute/core/Types.h"
+
 #include "tests/validation/Helpers.h"
 
 namespace arm_compute
@@ -40,7 +41,7 @@
     ARM_COMPUTE_UNUSED(oq_info);
 
     // Create reference
-    SimpleTensor<T> dst{ src.shape(), src.data_type(), 1 };
+    SimpleTensor<T> dst{src.shape(), src.data_type(), 1};
 
     // Compute reference
     const T a(info.a());
@@ -48,7 +49,7 @@
 #if defined(_OPENMP)
     #pragma omp parallel for
 #endif /* _OPENMP */
-    for(int i = 0; i < src.num_elements(); ++i)
+    for (int i = 0; i < src.num_elements(); ++i)
     {
         dst[i] = activate_float<T>(src[i], a, b, info.activation());
     }
@@ -57,7 +58,8 @@
 }
 
 template <>
-SimpleTensor<uint8_t> activation_layer<uint8_t>(const SimpleTensor<uint8_t> &src, ActivationLayerInfo info, const QuantizationInfo &oq_info)
+SimpleTensor<uint8_t>
+activation_layer<uint8_t>(const SimpleTensor<uint8_t> &src, ActivationLayerInfo info, const QuantizationInfo &oq_info)
 {
     const QuantizationInfo dst_qinfo = oq_info.empty() ? src.quantization_info() : oq_info;
 
@@ -68,7 +70,8 @@
 }
 
 template <>
-SimpleTensor<int8_t> activation_layer<int8_t>(const SimpleTensor<int8_t> &src, ActivationLayerInfo info, const QuantizationInfo &oq_info)
+SimpleTensor<int8_t>
+activation_layer<int8_t>(const SimpleTensor<int8_t> &src, ActivationLayerInfo info, const QuantizationInfo &oq_info)
 {
     const QuantizationInfo dst_qinfo = oq_info.empty() ? src.quantization_info() : oq_info;
 
@@ -79,7 +82,8 @@
 }
 
 template <>
-SimpleTensor<int16_t> activation_layer<int16_t>(const SimpleTensor<int16_t> &src, ActivationLayerInfo info, const QuantizationInfo &oq_info)
+SimpleTensor<int16_t>
+activation_layer<int16_t>(const SimpleTensor<int16_t> &src, ActivationLayerInfo info, const QuantizationInfo &oq_info)
 {
     const QuantizationInfo dst_qinfo = oq_info.empty() ? src.quantization_info() : oq_info;
 
@@ -88,9 +92,14 @@
     SimpleTensor<int16_t> dst     = convert_to_symmetric<int16_t>(dst_tmp, dst_qinfo);
     return dst;
 }
-template SimpleTensor<int32_t> activation_layer(const SimpleTensor<int32_t> &src, ActivationLayerInfo info, const QuantizationInfo &oq_info);
-template SimpleTensor<float> activation_layer(const SimpleTensor<float> &src, ActivationLayerInfo info, const QuantizationInfo &oq_info);
-template SimpleTensor<half> activation_layer(const SimpleTensor<half> &src, ActivationLayerInfo info, const QuantizationInfo &oq_info);
+template SimpleTensor<int32_t>
+activation_layer(const SimpleTensor<int32_t> &src, ActivationLayerInfo info, const QuantizationInfo &oq_info);
+template SimpleTensor<float>
+activation_layer(const SimpleTensor<float> &src, ActivationLayerInfo info, const QuantizationInfo &oq_info);
+template SimpleTensor<half>
+activation_layer(const SimpleTensor<half> &src, ActivationLayerInfo info, const QuantizationInfo &oq_info);
+template SimpleTensor<bfloat16>
+activation_layer(const SimpleTensor<bfloat16> &src, ActivationLayerInfo info, const QuantizationInfo &oq_info);
 } // namespace reference
 } // namespace validation
 } // namespace test