commit 03b2971ac69a86f10a1566938d1a25afee15746c
author Viet-Hoa Do <viet-hoa.do@arm.com> Wed Jun 01 11:47:14 2022 +0100
committer Viet-Hoa Do <viet-hoa.do@arm.com> Mon Nov 28 16:57:42 2022 +0000
tree aec7cfc047e1da278b4b71a706cda7b1b0faa158
parent 7dc0234331f2150a6b4ac5c2b49de419870f7cf5

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>

src/core/NEON/kernels/arm_conv/pooling/pooling_fp16.cpp

diff --git a/src/core/NEON/kernels/arm_conv/pooling/pooling_fp16.cpp b/src/core/NEON/kernels/arm_conv/pooling/pooling_fp16.cpp
index e3ce652..a7f3dd3a 100644
--- a/src/core/NEON/kernels/arm_conv/pooling/pooling_fp16.cpp
+++ b/src/core/NEON/kernels/arm_conv/pooling/pooling_fp16.cpp
@@ -33,6 +33,12 @@
 
 #include "kernels/cpp_nhwc_1x1_stride_any_depthfirst.hpp"
 #if defined(__aarch64__)
+#if defined(ARM_COMPUTE_ENABLE_SME)
+#include "kernels/sme_fp16_nhwc_max_2x2_s1_output2x2_depthfirst.hpp"
+#include "kernels/sme_fp16_nhwc_avg_3x3_s1_output2x2_depthfirst.hpp"
+#include "kernels/sme_fp16_nhwc_avg_generic_depthfirst.hpp"
+#include "kernels/sme_fp16_nhwc_max_generic_depthfirst.hpp"
+#endif  // defined(ARM_COMPUTE_ENABLE_SME)
 #if defined(ARM_COMPUTE_ENABLE_SVE)
 #include "kernels/sve_fp16_nhwc_max_2x2_s1_output2x2_depthfirst.hpp"
 #include "kernels/sve_fp16_nhwc_avg_3x3_s1_output2x2_depthfirst.hpp"
@@ -62,6 +68,58 @@
     },
   },
 #if defined(__aarch64__)
+#if defined(ARM_COMPUTE_ENABLE_SME)
+  {
+    PoolingMethod::DEPTHFIRST,
+    "sme_fp16_nhwc_max_2x2_s1_output2x2_depthfirst",
+    [] (const PoolingArgs &args, const Nothing &os) -> bool {
+      return args.cpu_info->has_sme() &&
+             is_supported<sme_fp16_nhwc_max_2x2_s1_output2x2_depthfirst>(args, os);
+    },
+    nullptr,
+    [] (const PoolingArgs &args, const Nothing &) -> PoolingCommon<__fp16, __fp16> * {
+      auto strat = new sme_fp16_nhwc_max_2x2_s1_output2x2_depthfirst(args.cpu_info);
+      return new PoolingDepthfirst<__fp16>(strat, args);
+    },
+  },
+  {
+    PoolingMethod::DEPTHFIRST,
+    "sme_fp16_nhwc_avg_3x3_s1_output2x2_depthfirst",
+    [] (const PoolingArgs &args, const Nothing &os) -> bool {
+      return args.cpu_info->has_sme() &&
+             is_supported<sme_fp16_nhwc_avg_3x3_s1_output2x2_depthfirst>(args, os);
+    },
+    nullptr,
+    [] (const PoolingArgs &args, const Nothing &) -> PoolingCommon<__fp16, __fp16> * {
+      auto strat = new sme_fp16_nhwc_avg_3x3_s1_output2x2_depthfirst(args.cpu_info);
+      return new PoolingDepthfirst<__fp16>(strat, args);
+    },
+  },
+  {
+    PoolingMethod::DEPTHFIRST,
+    "sme_fp16_nhwc_avg_generic_depthfirst",
+    [] (const PoolingArgs &args, const Nothing &) -> bool {
+      return args.cpu_info->has_sme() && args.pool_type == PoolingType::AVERAGE;
+    },
+    nullptr,
+    [] (const PoolingArgs &args, const Nothing &) -> PoolingCommon<__fp16, __fp16> * {
+      auto strat = new sme_fp16_nhwc_avg_generic_depthfirst(args.cpu_info);
+      return new PoolingDepthfirstGeneric<__fp16>(strat, args);
+    },
+  },
+  {
+    PoolingMethod::DEPTHFIRST,
+    "sme_fp16_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<__fp16, __fp16> * {
+      auto strat = new sme_fp16_nhwc_max_generic_depthfirst(args.cpu_info);
+      return new PoolingDepthfirstGeneric<__fp16>(strat, args);
+    },
+  },
+#endif  // defined(ARM_COMPUTE_ENABLE_SME)
 #if defined(ARM_COMPUTE_ENABLE_SVE)
   {
     PoolingMethod::DEPTHFIRST,