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, 21 insertions, 5 deletions

diff --git a/src/core/NEON/kernels/arm_gemm/kernels/a64_gemm_s8_8x12.hpp b/src/core/NEON/kernels/arm_gemm/kernels/a64_gemm_s8_8x12.hpp
index 7c7b894b08..83ccb4681b 100644
--- a/src/core/NEON/kernels/arm_gemm/kernels/a64_gemm_s8_8x12.hpp
+++ b/src/core/NEON/kernels/arm_gemm/kernels/a64_gemm_s8_8x12.hpp
@@ -61,13 +61,29 @@ public:
     StdTransformsFixed<operand_type, result_type, 8, 12, 4> transforms = {};
     StdTransformsFixed<operand_type, result_type, 8, 12, 4, true> transforms_quantized = {};
 
+    template<typename T>
     static PerformanceParameters get_performance_parameters(const CPUInfo *ci) {
-        switch (ci->get_cpu_model()) {
-            case CPUModel::A55r1:
-                return { 15.361, 0.9341, 0.1636 };
+        if (std::is_same<T, int8_t>::value) {
+            switch (ci->get_cpu_model()) {
+                case CPUModel::A510:
+                    return { 19.73, 2.81, 0.27 };
 
-            default:
-                return { 29.0698, 3.9793, 0.4003 };
+                case CPUModel::A55r1:
+                    return { 15.361, 0.9341, 0.1636 };
+
+                default:
+                    return { 29.0698, 3.9793, 0.4003 };
+            }
+        }
+
+        if (std::is_same<T, int32_t>::value) {
+            switch (ci->get_cpu_model()) {
+                case CPUModel::A510:
+                    return { 19.73, 3.41, 3.70 };
+
+                default:
+                    return { 31.81, 3.68, 8.01 };
+            }
         }
     }