From 491f30c0fff416007d97f4a5a043923861ef7b64 Mon Sep 17 00:00:00 2001
From: Gian Marco Iodice <gianmarco.iodice@arm.com>
Date: Mon, 2 Nov 2020 15:43:57 +0000
Subject: COMPMID-3939: Update GEMM heuristic Mali-G77

- Update heuristic for GEMM reshaped RHS only
- Fix left-over block size in CLGEMMMatrixMultiplyReshapedOlyRHSKernel

Change-Id: I34c738821ed2e4a537da4a15058eec164cb6b61f
Signed-off-by: Gian Marco Iodice <gianmarco.iodice@arm.com>
Reviewed-on: https://review.mlplatform.org/c/ml/ComputeLibrary/+/4305
Tested-by: Arm Jenkins <bsgcomp@arm.com>
Reviewed-by: Georgios Pinitas <georgios.pinitas@arm.com>
Comments-Addressed: Arm Jenkins <bsgcomp@arm.com>
---
 .../CLGEMMReshapedKernelConfigurationValhall.cpp   | 81 +++++++++++++++++++++-
 1 file changed, 78 insertions(+), 3 deletions(-)

(limited to 'src/core/CL/gemm/reshaped/CLGEMMReshapedKernelConfigurationValhall.cpp')

diff --git a/src/core/CL/gemm/reshaped/CLGEMMReshapedKernelConfigurationValhall.cpp b/src/core/CL/gemm/reshaped/CLGEMMReshapedKernelConfigurationValhall.cpp
index 519e903a5a..3f82dcab00 100644
--- a/src/core/CL/gemm/reshaped/CLGEMMReshapedKernelConfigurationValhall.cpp
+++ b/src/core/CL/gemm/reshaped/CLGEMMReshapedKernelConfigurationValhall.cpp
@@ -90,13 +90,88 @@ std::pair<GEMMLHSMatrixInfo, GEMMRHSMatrixInfo> CLGEMMReshapedKernelConfiguratio
     ARM_COMPUTE_UNUSED(k);
     ARM_COMPUTE_UNUSED(b);
 
-    if(n <= 4)
+    const float r_mn     = static_cast<float>(m) / static_cast<float>(n);
+    const float workload = (static_cast<float>(m) * static_cast<float>(n) * static_cast<float>(b)) / 20.0f;
+    const float r_mk = static_cast<float>(m) / static_cast<float>(k);
+    const float r_nk = static_cast<float>(n) / static_cast<float>(k);
+
+    if(r_mk <= 0.11824845522642136)
     {
-        return configure_lhs_rhs_info(m, n, 4, 2, 8, 8, 2, true, true, true, false);
+        if(workload <= 880.0)
+        {
+            return configure_lhs_rhs_info(m, n, 2, 4, 4, 1, 4, false, false, true, false, false);
+        }
+        else
+        {
+            if(r_nk <= 0.42521367967128754)
+            {
+                if(workload <= 1726.4000244140625)
+                {
+                    return configure_lhs_rhs_info(m, n, 4, 4, 4, 2, 2, false, false, true, false, false);
+                }
+                else
+                {
+                    return configure_lhs_rhs_info(m, n, 4, 4, 4, 2, 1, false, true, true, false, true);
+                }
+            }
+            else
+            {
+                if(workload <= 1241.6000366210938)
+                {
+                    return configure_lhs_rhs_info(m, n, 2, 4, 4, 1, 4, false, false, true, false, false);
+                }
+                else
+                {
+                    return configure_lhs_rhs_info(m, n, 4, 4, 4, 4, 4, false, false, true, false, false);
+                }
+            }
+        }
     }
     else
     {
-        return configure_lhs_rhs_info(m, n, 4, 8, 4, 4, 2, true, true, true, false);
+        if(workload <= 11404.7998046875)
+        {
+            if(r_mk <= 1.0126488208770752)
+            {
+                if(r_mn <= 2.545312523841858)
+                {
+                    return configure_lhs_rhs_info(m, n, 4, 4, 4, 2, 1, false, true, true, false, true);
+                }
+                else
+                {
+                    return configure_lhs_rhs_info(m, n, 2, 4, 4, 1, 4, false, false, true, false, false);
+                }
+            }
+            else
+            {
+                if(workload <= 2881.199951171875)
+                {
+                    return configure_lhs_rhs_info(m, n, 4, 4, 4, 4, 2, false, false, true, false, true);
+                }
+                else
+                {
+                    return configure_lhs_rhs_info(m, n, 4, 4, 4, 2, 1, false, true, true, false, true);
+                }
+            }
+        }
+        else
+        {
+            if(r_nk <= 0.5765306055545807)
+            {
+                if(r_mn <= 6.010416746139526)
+                {
+                    return configure_lhs_rhs_info(m, n, 4, 4, 4, 2, 1, false, true, true, false, true);
+                }
+                else
+                {
+                    return configure_lhs_rhs_info(m, n, 4, 4, 4, 2, 1, true, false, true, false, true);
+                }
+            }
+            else
+            {
+                return configure_lhs_rhs_info(m, n, 4, 4, 4, 2, 1, true, false, true, false, true);
+            }
+        }
     }
 }
 
-- 
cgit v1.2.1