From c6af9db1bb6b7a2093205f6f3088143bad660ff9 Mon Sep 17 00:00:00 2001
From: Gian Marco Iodice <gianmarco.iodice@arm.com>
Date: Thu, 24 Sep 2020 13:55:20 +0100
Subject: COMPMID-3673: Update heuristic for GEMM FP16 on Mali G76

- Update heuristic for CLGEMMReshapedKernel - FP16
- Update heuristic for CLGEMMReshapedOnlyRHSKernel - FP16

Change-Id: I35aa73e59d8c2d1bc6b2dd318fd8eeb3e42c27a4
Signed-off-by: Gian Marco Iodice <gianmarco.iodice@arm.com>
Reviewed-on: https://review.mlplatform.org/c/ml/ComputeLibrary/+/4026
Tested-by: Arm Jenkins <bsgcomp@arm.com>
Reviewed-by: SiCong Li <sicong.li@arm.com>
Comments-Addressed: Arm Jenkins <bsgcomp@arm.com>
---
 .../CLGEMMReshapedKernelConfigurationBifrost.cpp   | 87 +++++++++++++++++++++-
 1 file changed, 83 insertions(+), 4 deletions(-)

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

diff --git a/src/core/CL/gemm/reshaped/CLGEMMReshapedKernelConfigurationBifrost.cpp b/src/core/CL/gemm/reshaped/CLGEMMReshapedKernelConfigurationBifrost.cpp
index a533f14d02..b5fc074fb4 100644
--- a/src/core/CL/gemm/reshaped/CLGEMMReshapedKernelConfigurationBifrost.cpp
+++ b/src/core/CL/gemm/reshaped/CLGEMMReshapedKernelConfigurationBifrost.cpp
@@ -206,15 +206,94 @@ std::pair<GEMMLHSMatrixInfo, GEMMRHSMatrixInfo> CLGEMMReshapedKernelConfiguratio
 std::pair<GEMMLHSMatrixInfo, GEMMRHSMatrixInfo> CLGEMMReshapedKernelConfigurationBifrost::configure_G76_f16(unsigned int m, unsigned int n, unsigned int k, unsigned int b)
 {
     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;
+
+    if(workload <= 1049.59f)
     {
-        return configure_lhs_rhs_info(m, n, 4, 4, 4, 8, 2, true, true, true, false);
+        if(b <= 5)
+        {
+            if(workload <= 790.39f)
+            {
+                return configure_lhs_rhs_info(m,n,2,4,4,2,2,false,false,true,false,false);
+            }
+            else
+            {
+                if(workload <= 982.39f)
+                {
+                    return configure_lhs_rhs_info(m,n,4,2,4,4,4,false,false,true,false,false);
+                }
+                else
+                {
+                    return configure_lhs_rhs_info(m,n,2,4,4,2,1,false,true,true,false,false);
+                }
+            }
+        }
+        else
+        {
+            if(r_mn <= 0.21f)
+            {
+                if(r_mn <= 0.11f)
+                {
+                    return configure_lhs_rhs_info(m,n,2,4,4,2,2,false,false,true,false,false);
+                }
+                else
+                {
+                    return configure_lhs_rhs_info(m,n,4,4,4,4,4,false,true,true,false,false);
+                }
+            }
+            else
+            {
+                return configure_lhs_rhs_info(m,n,2,4,4,2,2,false,false,true,false,false);
+            }
+        }
     }
     else
     {
-        return configure_lhs_rhs_info(m, n, 4, 4, 4, 4, 8, true, true, true, false);
+        if(n <= 200)
+        {
+            if(workload <= 29772.79f)
+            {
+                if(m <= 64.5)
+                {
+                    return configure_lhs_rhs_info(m,n,4,4,4,2,4,true,false,true,false,false);
+                }
+                else
+                {
+                    return configure_lhs_rhs_info(m,n,4,4,4,2,2,false,true,true,false,false);
+                }
+            }
+            else
+            {
+                if(r_mn <= 1.09f)
+                {
+                    return configure_lhs_rhs_info(m,n,4,4,4,4,4,false,true,true,false,false);
+                }
+                else
+                {
+                    return configure_lhs_rhs_info(m,n,4,4,4,2,2,true,true,true,false,false);
+                }
+            }
+        }
+        else
+        {
+            if(m <= 43)
+            {
+                return configure_lhs_rhs_info(m,n,4,4,4,2,4,true,false,true,false,false);
+            }
+            else
+            {
+                if(workload <= 26364.79f)
+                {
+                    return configure_lhs_rhs_info(m,n,4,4,4,2,2,false,true,true,false,false);
+                }
+                else
+                {
+                    return configure_lhs_rhs_info(m,n,4,4,4,4,4,false,true,true,false,false);
+                }
+            }
+        }
     }
 }
 
-- 
cgit v1.2.1