From 03b2971ac69a86f10a1566938d1a25afee15746c Mon Sep 17 00:00:00 2001
From: Viet-Hoa Do <viet-hoa.do@arm.com>
Date: Wed, 1 Jun 2022 11:47:14 +0100
Subject: 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>
---
 .../NEON/kernels/arm_conv/pooling/pooling_s8q.cpp  | 30 ++++++++++++++++++++++
 1 file changed, 30 insertions(+)

(limited to 'src/core/NEON/kernels/arm_conv/pooling/pooling_s8q.cpp')

diff --git a/src/core/NEON/kernels/arm_conv/pooling/pooling_s8q.cpp b/src/core/NEON/kernels/arm_conv/pooling/pooling_s8q.cpp
index 6209f7cf2f..dcb3c8f57c 100644
--- a/src/core/NEON/kernels/arm_conv/pooling/pooling_s8q.cpp
+++ b/src/core/NEON/kernels/arm_conv/pooling/pooling_s8q.cpp
@@ -28,6 +28,10 @@
 #include "pooling_depthfirst_generic.hpp"
 
 #if defined(__aarch64__)
+#if defined(ARM_COMPUTE_ENABLE_SME)
+#include "kernels/sme_s8q_nhwc_avg_generic_depthfirst.hpp"
+#include "kernels/sme_s8q_nhwc_max_generic_depthfirst.hpp"
+#endif  // defined(ARM_COMPUTE_ENABLE_SME)
 #if defined(ARM_COMPUTE_ENABLE_SVE)
 #include "kernels/sve_s8q_nhwc_avg_generic_depthfirst.hpp"
 #include "kernels/sve_s8q_nhwc_max_generic_depthfirst.hpp"
@@ -43,6 +47,32 @@ namespace pooling {
 
 static const PoolingImplementation<int8_t, int8_t, Requantize32> pooling_s8q_methods[] = {
 #if defined(__aarch64__)
+#if defined(ARM_COMPUTE_ENABLE_SME)
+  {
+    PoolingMethod::DEPTHFIRST,
+    "sme_s8q_nhwc_avg_generic_depthfirst",
+    [] (const PoolingArgs &args, const Requantize32 &) -> bool {
+      return args.cpu_info->has_sme2() && args.pool_type == PoolingType::AVERAGE;
+    },
+    nullptr,
+    [] (const PoolingArgs &args, const Requantize32 &rq) -> PoolingCommon<int8_t, int8_t> * {
+      auto strat = new sme_s8q_nhwc_avg_generic_depthfirst(args.cpu_info);
+      return new PoolingDepthfirstGeneric<int8_t, int8_t, Requantize32>(strat, args, rq);
+    },
+  },
+  {
+    PoolingMethod::DEPTHFIRST,
+    "sme_s8q_nhwc_max_generic_depthfirst",
+    [] (const PoolingArgs &args, const Requantize32 &) -> bool {
+      return args.cpu_info->has_sme2() && args.pool_type == PoolingType::MAX;
+    },
+    nullptr,
+    [] (const PoolingArgs &args, const Requantize32 &rq) -> PoolingCommon<int8_t, int8_t> * {
+      auto strat = new sme_s8q_nhwc_max_generic_depthfirst(args.cpu_info);
+      return new PoolingDepthfirstGeneric<int8_t, int8_t, Requantize32>(strat, args, rq);
+    },
+  },
+#endif  // defined(ARM_COMPUTE_ENABLE_SME)
 #if defined(ARM_COMPUTE_ENABLE_SVE)
   {
     PoolingMethod::DEPTHFIRST,
-- 
cgit v1.2.1