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, 24 insertions, 9 deletions

diff --git a/src/core/NEON/kernels/arm_gemm/kernels/a64_sgemm_8x12.hpp b/src/core/NEON/kernels/arm_gemm/kernels/a64_sgemm_8x12.hpp
index f327e84861..39399e0ba8 100644
--- a/src/core/NEON/kernels/arm_gemm/kernels/a64_sgemm_8x12.hpp
+++ b/src/core/NEON/kernels/arm_gemm/kernels/a64_sgemm_8x12.hpp
@@ -28,6 +28,8 @@
 #include "../std_transforms_fixed.hpp"
 #include "../performance_parameters.hpp"
 
+#include "../bfloat.hpp"
+
 namespace arm_gemm {
 
 // Actual kernel implementations
@@ -68,19 +70,32 @@ public:
     // Use the standard fixed size transforms.
     StdTransformsFixed<operand_type, result_type, 8, 12> transforms = {};
 
+    template<typename T>
     static PerformanceParameters get_performance_parameters(const CPUInfo *ci) {
-        switch (ci->get_cpu_model()) {
-            case CPUModel::A55r1:
-                return { 3.954, 1.252, 1.141 };
+        if (std::is_same<T, float>::value) {
+            switch (ci->get_cpu_model()) {
+                case CPUModel::A55r1:
+                    return { 3.954, 1.252, 1.141 };
 
-            case CPUModel::A53:
-                return { 2.777, 0.987, 0.898 };
+                case CPUModel::A53:
+                    return { 2.777, 0.987, 0.898 };
 
-            case CPUModel::A73:
-                return { 2.885, 1.429, 1.163 };
+                case CPUModel::A73:
+                    return { 2.885, 1.429, 1.163 };
 
-            default:
-                return { 7.2307, 3.876, 2.932 };
+                default:
+                    return { 7.2307, 3.876, 2.932 };
+            }
+        }
+
+        if (std::is_same<T, bfloat16>::value) {
+            switch(ci->get_cpu_model()) {
+                case CPUModel::A510:
+                    return { 4.98, 2.27, 3.05 };
+
+                default:
+                    return { 7.99, 5.06, 7.32 };
+            }
         }
     }