Update GEMM assembly kernels

- Introduce Fp32 kernels with internal calculations in Bfloat16 when
fast_mode is enabled
- Improve kernel selection heuristics

Signed-off-by: Georgios Pinitas <georgios.pinitas@arm.com>
Change-Id: I68a9e7e862b6fd2721b46e0d7cc791091c4ab279
Reviewed-on: https://review.mlplatform.org/c/ml/ComputeLibrary/+/5965
Tested-by: Arm Jenkins <bsgcomp@arm.com>
Comments-Addressed: Arm Jenkins <bsgcomp@arm.com>

1 files changed, 14 insertions, 1 deletions

diff --git a/src/core/NEON/kernels/arm_gemm/quantize_wrapper.hpp b/src/core/NEON/kernels/arm_gemm/quantize_wrapper.hpp
index fdb4f584d8..1e2a9acc1d 100644
--- a/src/core/NEON/kernels/arm_gemm/quantize_wrapper.hpp
+++ b/src/core/NEON/kernels/arm_gemm/quantize_wrapper.hpp
@@ -1,5 +1,5 @@
 /*
- * Copyright (c) 2019-2020 Arm Limited.
+ * Copyright (c) 2019-2021 Arm Limited.
  *
  * SPDX-License-Identifier: MIT
  *
@@ -198,6 +198,19 @@ public:
         _params.bias = bias;
         _params.bias_multi_stride = bias_multi_stride;
     }
+
+    GemmConfig get_config() override {
+        GemmConfig c = _subgemm->get_config();
+
+        std::string n = "quantize_wrapper[";
+        n.append(c.filter);
+        n.append("]");
+
+        c.method = GemmMethod::QUANTIZE_WRAPPER;
+        c.filter = n;
+
+        return c;
+    }
 };
 
 } // namespace arm_gemm