Add Mali™-G720 and Mali™-G620 as GpuTargets

This patch adds adds the latest Gpus as Gpu Target and sets up kernel selection heuristics for MatMul to address some nightly issues.

Resolves: COMPMID-6766

Change-Id: I29dbb08c5ecfb3fcd63230b0b1675ab557074aca
Signed-off-by: Gunes Bayir <gunes.bayir@arm.com>
Reviewed-on: https://review.mlplatform.org/c/ml/ComputeLibrary/+/10902
Tested-by: Arm Jenkins <bsgcomp@arm.com>
Comments-Addressed: Arm Jenkins <bsgcomp@arm.com>
Reviewed-by: Jakub Sujak <jakub.sujak@arm.com>
Benchmark: Arm Jenkins <bsgcomp@arm.com>

12 files changed, 402 insertions, 26 deletions

diff --git a/src/runtime/CL/gemm/CLGEMMKernelSelection.h b/src/runtime/CL/gemm/CLGEMMKernelSelection.h
index c528dbcac4..98dd44b1bf 100644
--- a/src/runtime/CL/gemm/CLGEMMKernelSelection.h
+++ b/src/runtime/CL/gemm/CLGEMMKernelSelection.h
@@ -1,5 +1,5 @@
 /*
- * Copyright (c) 2020 Arm Limited.
+ * Copyright (c) 2020, 2023 Arm Limited.
  *
  * SPDX-License-Identifier: MIT
  *
@@ -21,8 +21,8 @@
  * OUT OF OR IN CONNECTION WITH THE SOFTWARE OR THE USE OR OTHER DEALINGS IN THE
  * SOFTWARE.
  */
-#ifndef SRC_CLGEMMKERNELSELECTION_H
-#define SRC_CLGEMMKERNELSELECTION_H
+#ifndef ACL_SRC_RUNTIME_CL_GEMM_CLGEMMKERNELSELECTION_H
+#define ACL_SRC_RUNTIME_CL_GEMM_CLGEMMKERNELSELECTION_H
 
 #include "arm_compute/runtime/CL/ICLGEMMKernelSelection.h"
 
@@ -53,6 +53,7 @@ public:
             case GPUTarget::BIFROST:
                 return std::make_unique<CLGEMMDefaultTypeBifrost>(gpu);
             case GPUTarget::VALHALL:
+            case GPUTarget::FIFTHGEN:
                 return std::make_unique<CLGEMMDefaultTypeValhall>(gpu);
             default:
                 ARM_COMPUTE_ERROR("Not supported GPU target");
@@ -61,4 +62,4 @@ public:
 };
 } // namespace cl_gemm
 } // namespace arm_compute
-#endif /* SRC_CLGEMMKERNELSELECTION_H */
+#endif // ACL_SRC_RUNTIME_CL_GEMM_CLGEMMKERNELSELECTION_H
diff --git a/src/runtime/heuristics/direct_conv/ClDirectConvKernelConfig.h b/src/runtime/heuristics/direct_conv/ClDirectConvKernelConfig.h
index 2c2509f70b..215b17ef79 100644
--- a/src/runtime/heuristics/direct_conv/ClDirectConvKernelConfig.h
+++ b/src/runtime/heuristics/direct_conv/ClDirectConvKernelConfig.h
@@ -1,5 +1,5 @@
 /*
- * Copyright (c) 2022 Arm Limited.
+ * Copyright (c) 2022-2023 Arm Limited.
  *
  * SPDX-License-Identifier: MIT
  *
@@ -21,8 +21,8 @@
  * OUT OF OR IN CONNECTION WITH THE SOFTWARE OR THE USE OR OTHER DEALINGS IN THE
  * SOFTWARE.
  */
-#ifndef SRC_RUNTIME_HEURISTICS_DIRECT_CONV_CLDIRECTCONVKERNELCONFIG
-#define SRC_RUNTIME_HEURISTICS_DIRECT_CONV_CLDIRECTCONVKERNELCONFIG
+#ifndef ACL_SRC_RUNTIME_HEURISTICS_DIRECT_CONV_CLDIRECTCONVKERNELCONFIG_H
+#define ACL_SRC_RUNTIME_HEURISTICS_DIRECT_CONV_CLDIRECTCONVKERNELCONFIG_H
 
 #include "src/runtime/heuristics/direct_conv/ClDirectConvDefaultConfigBifrost.h"
 #include "src/runtime/heuristics/direct_conv/ClDirectConvDefaultConfigValhall.h"
@@ -53,6 +53,7 @@ public:
             case GPUTarget::BIFROST:
                 return std::make_unique<ClDirectConvDefaultConfigBifrost>(gpu);
             case GPUTarget::VALHALL:
+            case GPUTarget::FIFTHGEN:
                 return std::make_unique<ClDirectConvDefaultConfigValhall>(gpu);
             default:
                 ARM_COMPUTE_ERROR("Not supported GPU target");
@@ -61,4 +62,4 @@ public:
 };
 } // namespace cl_direct_conv
 } // namespace arm_compute
-#endif /* SRC_RUNTIME_HEURISTICS_DIRECT_CONV_CLDIRECTCONVKERNELCONFIG */
+#endif // ACL_SRC_RUNTIME_HEURISTICS_DIRECT_CONV_CLDIRECTCONVKERNELCONFIG_H
diff --git a/src/runtime/heuristics/dwc_native/ClDWCNativeKernelConfig.h b/src/runtime/heuristics/dwc_native/ClDWCNativeKernelConfig.h
index 49ce6ff479..031cf1859a 100644
--- a/src/runtime/heuristics/dwc_native/ClDWCNativeKernelConfig.h
+++ b/src/runtime/heuristics/dwc_native/ClDWCNativeKernelConfig.h
@@ -1,5 +1,5 @@
 /*
- * Copyright (c) 2022 Arm Limited.
+ * Copyright (c) 2022-2023 Arm Limited.
  *
  * SPDX-License-Identifier: MIT
  *
@@ -21,8 +21,8 @@
  * OUT OF OR IN CONNECTION WITH THE SOFTWARE OR THE USE OR OTHER DEALINGS IN THE
  * SOFTWARE.
  */
-#ifndef SRC_RUNTIME_HEURISTICS_DWC_NATIVE_CLDWCNATIVEKERNELCONFIG
-#define SRC_RUNTIME_HEURISTICS_DWC_NATIVE_CLDWCNATIVEKERNELCONFIG
+#ifndef ACL_SRC_RUNTIME_HEURISTICS_DWC_NATIVE_CLDWCNATIVEKERNELCONFIG_H
+#define ACL_SRC_RUNTIME_HEURISTICS_DWC_NATIVE_CLDWCNATIVEKERNELCONFIG_H
 
 #include "src/runtime/heuristics/dwc_native/ClDWCNativeDefaultConfigBifrost.h"
 #include "src/runtime/heuristics/dwc_native/ClDWCNativeDefaultConfigValhall.h"
@@ -54,6 +54,7 @@ public:
             case GPUTarget::BIFROST:
                 return std::make_unique<ClDWCNativeDefaultConfigBifrost>(gpu);
             case GPUTarget::VALHALL:
+            case GPUTarget::FIFTHGEN:
                 return std::make_unique<ClDWCNativeDefaultConfigValhall>(gpu);
             default:
                 ARM_COMPUTE_ERROR("Not supported GPU target");
@@ -62,4 +63,4 @@ public:
 };
 } // namespace cl_dwc
 } // namespace arm_compute
-#endif /* SRC_RUNTIME_HEURISTICS_DWC_NATIVE_CLDWCNATIVEKERNELCONFIG */
+#endif // ACL_SRC_RUNTIME_HEURISTICS_DWC_NATIVE_CLDWCNATIVEKERNELCONFIG_H
diff --git a/src/runtime/heuristics/indirect_conv/ClIndirectConvKernelConfig.h b/src/runtime/heuristics/indirect_conv/ClIndirectConvKernelConfig.h
index dd614e1f68..5e7ba6f8e9 100644
--- a/src/runtime/heuristics/indirect_conv/ClIndirectConvKernelConfig.h
+++ b/src/runtime/heuristics/indirect_conv/ClIndirectConvKernelConfig.h
@@ -1,5 +1,5 @@
 /*
- * Copyright (c) 2022 Arm Limited.
+ * Copyright (c) 2022-2023 Arm Limited.
  *
  * SPDX-License-Identifier: MIT
  *
@@ -21,8 +21,8 @@
  * OUT OF OR IN CONNECTION WITH THE SOFTWARE OR THE USE OR OTHER DEALINGS IN THE
  * SOFTWARE.
  */
-#ifndef SRC_RUNTIME_HEURISTICS_INDIRECT_CONV_CLINDIRECTCONVKERNELCONFIG
-#define SRC_RUNTIME_HEURISTICS_INDIRECT_CONV_CLINDIRECTCONVKERNELCONFIG
+#ifndef ACL_SRC_RUNTIME_HEURISTICS_INDIRECT_CONV_CLINDIRECTCONVKERNELCONFIG_H
+#define ACL_SRC_RUNTIME_HEURISTICS_INDIRECT_CONV_CLINDIRECTCONVKERNELCONFIG_H
 
 #include "src/runtime/heuristics/indirect_conv/ClIndirectConvDefaultConfigValhall.h"
 #include "src/runtime/heuristics/indirect_conv/IClIndirectConvKernelConfig.h"
@@ -50,6 +50,7 @@ public:
             case GPUTarget::MIDGARD:
             case GPUTarget::BIFROST:
             case GPUTarget::VALHALL:
+            case GPUTarget::FIFTHGEN:
                 return std::make_unique<ClIndirectConvDefaultConfigValhall>(gpu);
             default:
                 ARM_COMPUTE_ERROR("Not supported GPU target");
@@ -58,4 +59,4 @@ public:
 };
 } // namespace cl_indirect_conv
 } // namespace arm_compute
-#endif /* SRC_RUNTIME_HEURISTICS_INDIRECT_CONV_CLINDIRECTCONVKERNELCONFIG */
+#endif // ACL_SRC_RUNTIME_HEURISTICS_INDIRECT_CONV_CLINDIRECTCONVKERNELCONFIG_H
diff --git a/src/runtime/heuristics/matmul_native/ClMatMulNativeDefaultConfigValhall.cpp b/src/runtime/heuristics/matmul_native/ClMatMulNativeDefaultConfigValhall.cpp
index 4b923547c4..3a02a60650 100644
--- a/src/runtime/heuristics/matmul_native/ClMatMulNativeDefaultConfigValhall.cpp
+++ b/src/runtime/heuristics/matmul_native/ClMatMulNativeDefaultConfigValhall.cpp
@@ -62,6 +62,7 @@ ClMatMulNativeDefaultConfigValhall::configure(const ITensorInfo *lhs, const ITen
     switch (_target)
     {
         case GPUTarget::G715:
+        case GPUTarget::G615:
             func = configs_G715.get_function(lhs->data_type());
             break;
         case GPUTarget::G710:
diff --git a/src/runtime/heuristics/matmul_native/ClMatMulNativeDefaultVariantValhall.cpp b/src/runtime/heuristics/matmul_native/ClMatMulNativeDefaultVariantValhall.cpp
new file mode 100644
index 0000000000..3878f698fd
--- /dev/null
+++ b/src/runtime/heuristics/matmul_native/ClMatMulNativeDefaultVariantValhall.cpp
@@ -0,0 +1,113 @@
+/*
+ * Copyright (c) 2023 Arm Limited.
+ *
+ * SPDX-License-Identifier: MIT
+ *
+ * Permission is hereby granted, free of charge, to any person obtaining a copy
+ * of this software and associated documentation files (the "Software"), to
+ * deal in the Software without restriction, including without limitation the
+ * rights to use, copy, modify, merge, publish, distribute, sublicense, and/or
+ * sell copies of the Software, and to permit persons to whom the Software is
+ * furnished to do so, subject to the following conditions:
+ *
+ * The above copyright notice and this permission notice shall be included in all
+ * copies or substantial portions of the Software.
+ *
+ * THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS OR
+ * IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF MERCHANTABILITY,
+ * FITNESS FOR A PARTICULAR PURPOSE AND NONINFRINGEMENT. IN NO EVENT SHALL THE
+ * AUTHORS OR COPYRIGHT HOLDERS BE LIABLE FOR ANY CLAIM, DAMAGES OR OTHER
+ * LIABILITY, WHETHER IN AN ACTION OF CONTRACT, TORT OR OTHERWISE, ARISING FROM,
+ * OUT OF OR IN CONNECTION WITH THE SOFTWARE OR THE USE OR OTHER DEALINGS IN THE
+ * SOFTWARE.
+ */
+#include "src/runtime/heuristics/matmul_native/ClMatMulNativeDefaultVariantValhall.h"
+
+#include "arm_compute/core/Error.h"
+#include "arm_compute/core/GPUTarget.h"
+#include "arm_compute/core/TensorInfo.h"
+
+namespace arm_compute
+{
+namespace cl_matmul
+{
+ClMatMulNativeDefaultVariantValhall::ClMatMulNativeDefaultVariantValhall(GPUTarget gpu)
+    : IClMatMulNativeKernelVariant(gpu)
+{
+}
+
+MatMulKernelType ClMatMulNativeDefaultVariantValhall::select_kernel(const ITensorInfo         *lhs,
+                                                                    const ITensorInfo         *rhs,
+                                                                    const MatMulInfo          &info,
+                                                                    const ActivationLayerInfo &act_info)
+{
+    ARM_COMPUTE_UNUSED(rhs);
+
+    using VariantFunctionExecutorPtr =
+        MatMulKernelType (ClMatMulNativeDefaultVariantValhall::*)(int k, bool act_enabled);
+
+    ClMatMulNativeVariantArray<VariantFunctionExecutorPtr> configs_G715(
+        &ClMatMulNativeDefaultVariantValhall::configure_G715_float,
+        &ClMatMulNativeDefaultVariantValhall::configure_G715_quantized);
+
+    ClMatMulNativeVariantArray<VariantFunctionExecutorPtr> configs_default(
+        &ClMatMulNativeDefaultVariantValhall::configure_default_float,
+        &ClMatMulNativeDefaultVariantValhall::configure_default_quantized);
+
+    VariantFunctionExecutorPtr func = nullptr;
+    switch (_target)
+    {
+        case GPUTarget::G715:
+        case GPUTarget::G615:
+            func = configs_G715.get_function(lhs->data_type());
+            break;
+        default:
+            func = configs_default.get_function(lhs->data_type());
+            break;
+    }
+
+    const int  k           = info.adj_lhs() ? lhs->tensor_shape().y() : lhs->tensor_shape().x();
+    const bool act_enabled = act_info.enabled();
+
+    ARM_COMPUTE_ERROR_ON_MSG(func == nullptr, "Data type not supported for matmul native");
+    return (this->*func)(k, act_enabled);
+}
+
+MatMulKernelType ClMatMulNativeDefaultVariantValhall::configure_G715_float(int k, bool act_enabled)
+{
+    // MMUL kernel works only when K is a multiple of 4
+    if (!act_enabled && k % 4 == 0)
+    {
+        return MatMulKernelType::NATIVE_MMUL_FP;
+    }
+
+    return MatMulKernelType::NATIVE_FP;
+}
+
+MatMulKernelType ClMatMulNativeDefaultVariantValhall::configure_G715_quantized(int k, bool act_enabled)
+{
+    // MMUL kernel works only when K is a multiple of 16
+    if (!act_enabled && k % 16 == 0)
+    {
+        return MatMulKernelType::NATIVE_MMUL_QUANTIZED;
+    }
+
+    return MatMulKernelType::NATIVE_QUANTIZED;
+}
+
+MatMulKernelType ClMatMulNativeDefaultVariantValhall::configure_default_float(int k, bool act_enabled)
+{
+    ARM_COMPUTE_UNUSED(k, act_enabled);
+
+    return MatMulKernelType::NATIVE_FP;
+}
+
+MatMulKernelType ClMatMulNativeDefaultVariantValhall::configure_default_quantized(int k, bool act_enabled)
+{
+    ARM_COMPUTE_UNUSED(k, act_enabled);
+
+    return MatMulKernelType::NATIVE_QUANTIZED;
+}
+
+} // namespace cl_matmul
+} // namespace arm_compute
diff --git a/src/runtime/heuristics/matmul_native/ClMatMulNativeDefaultVariantValhall.h b/src/runtime/heuristics/matmul_native/ClMatMulNativeDefaultVariantValhall.h
new file mode 100644
index 0000000000..a202676e98
--- /dev/null
+++ b/src/runtime/heuristics/matmul_native/ClMatMulNativeDefaultVariantValhall.h
@@ -0,0 +1,57 @@
+/*
+ * Copyright (c) 2023 Arm Limited.
+ *
+ * SPDX-License-Identifier: MIT
+ *
+ * Permission is hereby granted, free of charge, to any person obtaining a copy
+ * of this software and associated documentation files (the "Software"), to
+ * deal in the Software without restriction, including without limitation the
+ * rights to use, copy, modify, merge, publish, distribute, sublicense, and/or
+ * sell copies of the Software, and to permit persons to whom the Software is
+ * furnished to do so, subject to the following conditions:
+ *
+ * The above copyright notice and this permission notice shall be included in all
+ * copies or substantial portions of the Software.
+ *
+ * THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS OR
+ * IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF MERCHANTABILITY,
+ * FITNESS FOR A PARTICULAR PURPOSE AND NONINFRINGEMENT. IN NO EVENT SHALL THE
+ * AUTHORS OR COPYRIGHT HOLDERS BE LIABLE FOR ANY CLAIM, DAMAGES OR OTHER
+ * LIABILITY, WHETHER IN AN ACTION OF CONTRACT, TORT OR OTHERWISE, ARISING FROM,
+ * OUT OF OR IN CONNECTION WITH THE SOFTWARE OR THE USE OR OTHER DEALINGS IN THE
+ * SOFTWARE.
+ */
+#ifndef ACL_SRC_RUNTIME_HEURISTICS_MATMUL_NATIVE_CLMATMULNATIVEDEFAULTVARIANTVALHALL_H
+#define ACL_SRC_RUNTIME_HEURISTICS_MATMUL_NATIVE_CLMATMULNATIVEDEFAULTVARIANTVALHALL_H
+
+#include "src/runtime/heuristics/matmul_native/IClMatMulNativeKernelVariant.h"
+
+namespace arm_compute
+{
+namespace cl_matmul
+{
+/** Valhall based OpenCL matmul configuration */
+class ClMatMulNativeDefaultVariantValhall final : public IClMatMulNativeKernelVariant
+{
+public:
+    /** Constructor
+     *
+     * @param[in] gpu GPU target
+     */
+    ClMatMulNativeDefaultVariantValhall(GPUTarget gpu);
+
+    // Inherited overridden method
+    MatMulKernelType select_kernel(const ITensorInfo         *lhs,
+                                   const ITensorInfo         *rhs,
+                                   const MatMulInfo          &info,
+                                   const ActivationLayerInfo &act_info) override;
+
+private:
+    MatMulKernelType configure_G715_float(int k, bool act_enabled);
+    MatMulKernelType configure_G715_quantized(int k, bool act_enabled);
+    MatMulKernelType configure_default_float(int k, bool act_enabled);
+    MatMulKernelType configure_default_quantized(int k, bool act_enabled);
+};
+} // namespace cl_matmul
+} // namespace arm_compute
+#endif // ACL_SRC_RUNTIME_HEURISTICS_MATMUL_NATIVE_CLMATMULNATIVEDEFAULTVARIANTVALHALL_H
diff --git a/src/runtime/heuristics/matmul_native/ClMatMulNativeHelpers.h b/src/runtime/heuristics/matmul_native/ClMatMulNativeHelpers.h
index a114fffa68..699f5fe8c1 100644
--- a/src/runtime/heuristics/matmul_native/ClMatMulNativeHelpers.h
+++ b/src/runtime/heuristics/matmul_native/ClMatMulNativeHelpers.h
@@ -21,8 +21,8 @@
  * OUT OF OR IN CONNECTION WITH THE SOFTWARE OR THE USE OR OTHER DEALINGS IN THE
  * SOFTWARE.
  */
-#ifndef SRC_RUNTIME_HEURISTICS_MATMUL_NATIVE_CLMATMULNATIVEHELPERS
-#define SRC_RUNTIME_HEURISTICS_MATMUL_NATIVE_CLMATMULNATIVEHELPERS
+#ifndef ACL_SRC_RUNTIME_HEURISTICS_MATMUL_NATIVE_CLMATMULNATIVEHELPERS_H
+#define ACL_SRC_RUNTIME_HEURISTICS_MATMUL_NATIVE_CLMATMULNATIVEHELPERS_H
 
 #include "arm_compute/core/Types.h"
 
@@ -80,4 +80,4 @@ MatMulKernelInfo find_info(const MatMulNativeConfigsMatrix &configs,
                            unsigned int                     b);
 } // namespace cl_matmul
 } // namespace arm_compute
-#endif /* SRC_RUNTIME_HEURISTICS_MATMUL_NATIVE_CLMATMULNATIVEHELPERS */
+#endif // ACL_SRC_RUNTIME_HEURISTICS_MATMUL_NATIVE_CLMATMULNATIVEHELPERS_H
diff --git a/src/runtime/heuristics/matmul_native/ClMatMulNativeKernelConfig.h b/src/runtime/heuristics/matmul_native/ClMatMulNativeKernelConfig.h
index b10018a6d2..e7485bca81 100644
--- a/src/runtime/heuristics/matmul_native/ClMatMulNativeKernelConfig.h
+++ b/src/runtime/heuristics/matmul_native/ClMatMulNativeKernelConfig.h
@@ -21,8 +21,8 @@
  * OUT OF OR IN CONNECTION WITH THE SOFTWARE OR THE USE OR OTHER DEALINGS IN THE
  * SOFTWARE.
  */
-#ifndef SRC_RUNTIME_HEURISTICS_MATMUL_NATIVE_CLMATMULNATIVEKERNELCONFIG
-#define SRC_RUNTIME_HEURISTICS_MATMUL_NATIVE_CLMATMULNATIVEKERNELCONFIG
+#ifndef ACL_SRC_RUNTIME_HEURISTICS_MATMUL_NATIVE_CLMATMULNATIVEKERNELCONFIG_H
+#define ACL_SRC_RUNTIME_HEURISTICS_MATMUL_NATIVE_CLMATMULNATIVEKERNELCONFIG_H
 
 #include "src/runtime/heuristics/matmul_native/ClMatMulNativeDefaultConfigValhall.h"
 #include "src/runtime/heuristics/matmul_native/IClMatMulNativeKernelConfig.h"
@@ -50,6 +50,7 @@ public:
             case GPUTarget::MIDGARD:
             case GPUTarget::BIFROST:
             case GPUTarget::VALHALL:
+            case GPUTarget::FIFTHGEN:
                 return std::make_unique<ClMatMulNativeDefaultConfigValhall>(gpu);
             default:
                 ARM_COMPUTE_ERROR("Not supported GPU target");
@@ -58,4 +59,4 @@ public:
 };
 } // namespace cl_matmul
 } // namespace arm_compute
-#endif /* SRC_RUNTIME_HEURISTICS_MATMUL_NATIVE_CLMATMULNATIVEKERNELCONFIG */
+#endif // ACL_SRC_RUNTIME_HEURISTICS_MATMUL_NATIVE_CLMATMULNATIVEKERNELCONFIG_H
diff --git a/src/runtime/heuristics/matmul_native/ClMatMulNativeKernelVariant.h b/src/runtime/heuristics/matmul_native/ClMatMulNativeKernelVariant.h
new file mode 100644
index 0000000000..c2895b8919
--- /dev/null
+++ b/src/runtime/heuristics/matmul_native/ClMatMulNativeKernelVariant.h
@@ -0,0 +1,63 @@
+/*
+ * Copyright (c) 2023 Arm Limited.
+ *
+ * SPDX-License-Identifier: MIT
+ *
+ * Permission is hereby granted, free of charge, to any person obtaining a copy
+ * of this software and associated documentation files (the "Software"), to
+ * deal in the Software without restriction, including without limitation the
+ * rights to use, copy, modify, merge, publish, distribute, sublicense, and/or
+ * sell copies of the Software, and to permit persons to whom the Software is
+ * furnished to do so, subject to the following conditions:
+ *
+ * The above copyright notice and this permission notice shall be included in all
+ * copies or substantial portions of the Software.
+ *
+ * THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS OR
+ * IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF MERCHANTABILITY,
+ * FITNESS FOR A PARTICULAR PURPOSE AND NONINFRINGEMENT. IN NO EVENT SHALL THE
+ * AUTHORS OR COPYRIGHT HOLDERS BE LIABLE FOR ANY CLAIM, DAMAGES OR OTHER
+ * LIABILITY, WHETHER IN AN ACTION OF CONTRACT, TORT OR OTHERWISE, ARISING FROM,
+ * OUT OF OR IN CONNECTION WITH THE SOFTWARE OR THE USE OR OTHER DEALINGS IN THE
+ * SOFTWARE.
+ */
+#ifndef ACL_SRC_RUNTIME_HEURISTICS_MATMUL_NATIVE_CLMATMULNATIVEKERNELVARIANT_H
+#define ACL_SRC_RUNTIME_HEURISTICS_MATMUL_NATIVE_CLMATMULNATIVEKERNELVARIANT_H
+
+#include "src/runtime/heuristics/matmul_native/ClMatMulNativeDefaultVariantValhall.h"
+#include "src/runtime/heuristics/matmul_native/IClMatMulNativeKernelVariant.h"
+
+#include <memory>
+
+namespace arm_compute
+{
+namespace cl_matmul
+{
+
+/** ClMatMul variant factory class */
+class ClMatMulNativeKernelVariantFactory final
+{
+public:
+    /** Static method to call the ClMatMul configuration class accordingly with the GPU target
+     *
+     * @param[in] gpu GPU target
+     *
+     * @return IClMatMulNativeKernelVariant
+     */
+    static std::unique_ptr<IClMatMulNativeKernelVariant> create(GPUTarget gpu)
+    {
+        switch (get_arch_from_target(gpu))
+        {
+            case GPUTarget::MIDGARD:
+            case GPUTarget::BIFROST:
+            case GPUTarget::VALHALL:
+            case GPUTarget::FIFTHGEN:
+                return std::make_unique<ClMatMulNativeDefaultVariantValhall>(gpu);
+            default:
+                ARM_COMPUTE_ERROR("Not supported GPU target");
+        }
+    }
+};
+} // namespace cl_matmul
+} // namespace arm_compute
+#endif // ACL_SRC_RUNTIME_HEURISTICS_MATMUL_NATIVE_CLMATMULNATIVEKERNELVARIANT_H
diff --git a/src/runtime/heuristics/matmul_native/IClMatMulNativeKernelConfig.h b/src/runtime/heuristics/matmul_native/IClMatMulNativeKernelConfig.h
index b9b091100c..00ba3641d5 100644
--- a/src/runtime/heuristics/matmul_native/IClMatMulNativeKernelConfig.h
+++ b/src/runtime/heuristics/matmul_native/IClMatMulNativeKernelConfig.h
@@ -21,8 +21,8 @@
  * OUT OF OR IN CONNECTION WITH THE SOFTWARE OR THE USE OR OTHER DEALINGS IN THE
  * SOFTWARE.
  */
-#ifndef SRC_RUNTIME_HEURISTICS_MATMUL_NATIVE_ICLMATMULNATIVEKERNELCONFIG
-#define SRC_RUNTIME_HEURISTICS_MATMUL_NATIVE_ICLMATMULNATIVEKERNELCONFIG
+#ifndef ACL_SRC_RUNTIME_HEURISTICS_MATMUL_NATIVE_ICLMATMULNATIVEKERNELCONFIG_H
+#define ACL_SRC_RUNTIME_HEURISTICS_MATMUL_NATIVE_ICLMATMULNATIVEKERNELCONFIG_H
 
 #include "arm_compute/core/GPUTarget.h"
 #include "arm_compute/core/KernelDescriptors.h"
@@ -85,7 +85,9 @@ private:
     std::array<T, 3> _configs;
 };
 
-/** Basic interface for the matmul native kernel configuration */
+/** Basic interface for the matmul native kernel configuration
+ *  This is the base class that chooses architecture specific kernel configurations.
+*/
 class IClMatMulNativeKernelConfig
 {
 public:
@@ -112,4 +114,4 @@ protected:
 };
 } // namespace cl_matmul
 } // namespace arm_compute
-#endif /* SRC_RUNTIME_HEURISTICS_MATMUL_NATIVE_ICLMATMULNATIVEKERNELCONFIG */
+#endif // ACL_SRC_RUNTIME_HEURISTICS_MATMUL_NATIVE_ICLMATMULNATIVEKERNELCONFIG_H
diff --git a/src/runtime/heuristics/matmul_native/IClMatMulNativeKernelVariant.h b/src/runtime/heuristics/matmul_native/IClMatMulNativeKernelVariant.h
new file mode 100644
index 0000000000..eac41dd6a3
--- /dev/null
+++ b/src/runtime/heuristics/matmul_native/IClMatMulNativeKernelVariant.h
@@ -0,0 +1,135 @@
+/*
+ * Copyright (c) 2023 Arm Limited.
+ *
+ * SPDX-License-Identifier: MIT
+ *
+ * Permission is hereby granted, free of charge, to any person obtaining a copy
+ * of this software and associated documentation files (the "Software"), to
+ * deal in the Software without restriction, including without limitation the
+ * rights to use, copy, modify, merge, publish, distribute, sublicense, and/or
+ * sell copies of the Software, and to permit persons to whom the Software is
+ * furnished to do so, subject to the following conditions:
+ *
+ * The above copyright notice and this permission notice shall be included in all
+ * copies or substantial portions of the Software.
+ *
+ * THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS OR
+ * IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF MERCHANTABILITY,
+ * FITNESS FOR A PARTICULAR PURPOSE AND NONINFRINGEMENT. IN NO EVENT SHALL THE
+ * AUTHORS OR COPYRIGHT HOLDERS BE LIABLE FOR ANY CLAIM, DAMAGES OR OTHER
+ * LIABILITY, WHETHER IN AN ACTION OF CONTRACT, TORT OR OTHERWISE, ARISING FROM,
+ * OUT OF OR IN CONNECTION WITH THE SOFTWARE OR THE USE OR OTHER DEALINGS IN THE
+ * SOFTWARE.
+ */
+#ifndef ACL_SRC_RUNTIME_HEURISTICS_MATMUL_NATIVE_ICLMATMULNATIVEKERNELVARIANT_H
+#define ACL_SRC_RUNTIME_HEURISTICS_MATMUL_NATIVE_ICLMATMULNATIVEKERNELVARIANT_H
+
+#include "arm_compute/core/CoreTypes.h" // DataType
+#include "arm_compute/core/GPUTarget.h"
+#include "arm_compute/core/ITensorInfo.h"
+#include "arm_compute/function_info/ActivationLayerInfo.h"
+#include "arm_compute/function_info/MatMulInfo.h"
+
+#include "src/core/common/Macros.h"
+
+#include <array>
+
+namespace arm_compute
+{
+namespace cl_matmul
+{
+enum class MatMulKernelType
+{
+    /** Native matrix multiplication for FP types */
+    NATIVE_FP,
+
+    /** Native matrix multiplication for quantized types */
+    NATIVE_QUANTIZED,
+
+    /** Native matrix multiplication using MMUL extension for FP types */
+    NATIVE_MMUL_FP,
+
+    /** Native matrix multiplication using MMUL extension for Quantized types */
+    NATIVE_MMUL_QUANTIZED
+};
+
+/** Basic container for the OpenCL MatMul Native variant functions */
+template <class T>
+class ClMatMulNativeVariantArray
+{
+public:
+    /** Alias for Float index */
+    static constexpr size_t DT_FLOAT = 0;
+    /** Alias for Quantized type index */
+    static constexpr size_t DT_QUANTIZED = 1;
+
+    /** Constructor
+     *
+     * @param[in] func_float     Function to call for matmul native float (F32, F16)
+     * @param[in] func_quantized Function to call for matmul native quantized (QASYMM8, QASYMM8_SIGNED, QSYMM8_PER_CHANNEL)
+     *
+     */
+    ClMatMulNativeVariantArray(T func_float, T func_quantized) : _configs{func_float, func_quantized}
+    {
+    }
+
+    /** Method to return the matmul native variant function based on data type
+     *
+     * @param[in] data_type Input data type
+     *
+     * @return the valid function otherwise it returns nullptr if the data type is not valid
+     */
+    T get_function(DataType data_type)
+    {
+        switch (data_type)
+        {
+            case DataType::F32:
+            case DataType::F16:
+                return _configs.at(DT_FLOAT);
+            case DataType::QASYMM8:
+            case DataType::QASYMM8_SIGNED:
+            case DataType::QSYMM8_PER_CHANNEL:
+                return _configs.at(DT_QUANTIZED);
+            default:
+                return nullptr;
+        }
+    }
+
+private:
+    std::array<T, 2> _configs;
+};
+
+/** Basic interface for the matmul native kernel variant
+ *  This is the base class that chooses architecture specific kernel variants.
+*/
+class IClMatMulNativeKernelVariant
+{
+public:
+    /** Constructor
+     *
+     * @param[in] arch GPU target
+     */
+    IClMatMulNativeKernelVariant(GPUTarget arch) : _target(arch)
+    {
+    }
+    ARM_COMPUTE_DISALLOW_COPY_ALLOW_MOVE(IClMatMulNativeKernelVariant);
+    /** Virtual destructor */
+    virtual ~IClMatMulNativeKernelVariant() = default;
+    /** This method returns the @ref MatMulKernelType for the given inputs
+     *
+     * @param[in] lhs      LHS tensor
+     * @param[in] rhs      RHS tensor
+     * @param[in] info     MatMul info
+     * @param[in] act_info Activation layer info
+     */
+    virtual MatMulKernelType select_kernel(const ITensorInfo         *lhs,
+                                           const ITensorInfo         *rhs,
+                                           const MatMulInfo          &info,
+                                           const ActivationLayerInfo &act_info) = 0;
+
+protected:
+    GPUTarget _target;
+};
+} // namespace cl_matmul
+} // namespace arm_compute
+#endif // ACL_SRC_RUNTIME_HEURISTICS_MATMUL_NATIVE_ICLMATMULNATIVEKERNELVARIANT_H