Integrate SME2 kernels

* Add SME/SME2 detection.
* Integrate SME2 implementation for:
  - Normal convolution
  - Winograd
  - Depthwise convolution
  - Pooling

Resolves: COMPMID-5700
Signed-off-by: Viet-Hoa Do <viet-hoa.do@arm.com>
Change-Id: I2f1ca1d05f8cfeee9309ed1c0a36096a4a6aad5c
Reviewed-on: https://review.mlplatform.org/c/ml/ComputeLibrary/+/8692
Reviewed-by: Gunes Bayir <gunes.bayir@arm.com>
Tested-by: Arm Jenkins <bsgcomp@arm.com>
Comments-Addressed: Arm Jenkins <bsgcomp@arm.com>

1 files changed, 44 insertions, 0 deletions

diff --git a/src/core/NEON/kernels/arm_conv/pooling/pooling_s8.cpp b/src/core/NEON/kernels/arm_conv/pooling/pooling_s8.cpp
index 0867abc054..8d08ddc43f 100644
--- a/src/core/NEON/kernels/arm_conv/pooling/pooling_s8.cpp
+++ b/src/core/NEON/kernels/arm_conv/pooling/pooling_s8.cpp
@@ -30,6 +30,11 @@
 
 #include "kernels/cpp_nhwc_1x1_stride_any_depthfirst.hpp"
 #if defined(__aarch64__)
+#if defined(ARM_COMPUTE_ENABLE_SME)
+#include "kernels/sme_s8_nhwc_avg_generic_depthfirst.hpp"
+#include "kernels/sme_s8_nhwc_max_2x2_s1_output2x2_depthfirst.hpp"
+#include "kernels/sme_s8_nhwc_max_generic_depthfirst.hpp"
+#endif  // defined(ARM_COMPUTE_ENABLE_SME)
 #if defined(ARM_COMPUTE_ENABLE_SVE)
 #include "kernels/sve_s8_nhwc_avg_generic_depthfirst.hpp"
 #include "kernels/sve_s8_nhwc_max_2x2_s1_output2x2_depthfirst.hpp"
@@ -59,6 +64,45 @@ static const PoolingImplementation<int8_t, int8_t> pooling_s8_methods[] = {
     },
   },
 #if defined(__aarch64__)
+#if defined(ARM_COMPUTE_ENABLE_SME)
+  {
+    PoolingMethod::DEPTHFIRST,
+    "sme_s8_nhwc_max_2x2_s1_output2x2_depthfirst",
+    [] (const PoolingArgs &args, const Nothing &os) -> bool {
+      return args.cpu_info->has_sme() &&
+             is_supported<sme_s8_nhwc_max_2x2_s1_output2x2_depthfirst>(args, os);
+    },
+    nullptr,
+    [] (const PoolingArgs &args, const Nothing &) -> PoolingCommon<int8_t, int8_t> * {
+      auto strat = new sme_s8_nhwc_max_2x2_s1_output2x2_depthfirst(args.cpu_info);
+      return new PoolingDepthfirst<int8_t>(strat, args);
+    },
+  },
+  {
+    PoolingMethod::DEPTHFIRST,
+    "sme_s8_nhwc_avg_generic_depthfirst",
+    [] (const PoolingArgs &args, const Nothing &) -> bool {
+      return args.cpu_info->has_sme2() && args.pool_type == PoolingType::AVERAGE;
+    },
+    nullptr,
+    [] (const PoolingArgs &args, const Nothing &) -> PoolingCommon<int8_t, int8_t> * {
+      auto strat = new sme_s8_nhwc_avg_generic_depthfirst(args.cpu_info);
+      return new PoolingDepthfirstGeneric<int8_t>(strat, args);
+    },
+  },
+  {
+    PoolingMethod::DEPTHFIRST,
+    "sme_s8_nhwc_max_generic_depthfirst",
+    [] (const PoolingArgs &args, const Nothing &) -> bool {
+      return args.cpu_info->has_sme() && args.pool_type == PoolingType::MAX;
+    },
+    nullptr,
+    [] (const PoolingArgs &args, const Nothing &) -> PoolingCommon<int8_t, int8_t> * {
+      auto strat = new sme_s8_nhwc_max_generic_depthfirst(args.cpu_info);
+      return new PoolingDepthfirstGeneric<int8_t>(strat, args);
+    },
+  },
+#endif  // defined(ARM_COMPUTE_ENABLE_SME)
 #if defined(ARM_COMPUTE_ENABLE_SVE)
   {
     PoolingMethod::DEPTHFIRST,