From 839e19865d4b654899d1da5cfb94304841e7f210 Mon Sep 17 00:00:00 2001
From: Gian Marco Iodice <gianmarco.iodice@arm.com>
Date: Thu, 29 Oct 2020 13:36:50 +0000
Subject: COMPMID-3930: Update CLGEMM heuristic for fp16. Mali-G76

- Since the GEMM kernel can now work without padding, the heuristic
  requires to be fine-tuned to exploit this feature
- The heuristic affects Mali-G76 FP16 only

Change-Id: Ia430627f02131ad956ce2219b80c83c8e7cabaf2
Signed-off-by: Gian Marco Iodice <gianmarco.iodice@arm.com>
Reviewed-on: https://review.mlplatform.org/c/ml/ComputeLibrary/+/4284
Tested-by: Arm Jenkins <bsgcomp@arm.com>
Comments-Addressed: Arm Jenkins <bsgcomp@arm.com>
Reviewed-by: Michele Di Giorgio <michele.digiorgio@arm.com>
Reviewed-by: SiCong Li <sicong.li@arm.com>
---
 .../CLGEMMReshapedKernelConfigurationBifrost.cpp   | 79 +++-------------------
 1 file changed, 11 insertions(+), 68 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 00c284facc..c1ca187a70 100644
--- a/src/core/CL/gemm/reshaped/CLGEMMReshapedKernelConfigurationBifrost.cpp
+++ b/src/core/CL/gemm/reshaped/CLGEMMReshapedKernelConfigurationBifrost.cpp
@@ -205,95 +205,38 @@ 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);
-
-    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(workload <= 1049.59f)
+    if(workload <= 1422.40f)
     {
-        if(b <= 5)
+        if(r_mk <= 2.45f)
         {
-            if(workload <= 790.39f)
+            if(workload <= 801.60f)
             {
-                return configure_lhs_rhs_info(m, n, 2, 4, 4, 2, 2, false, false, true, false, false);
+                return configure_lhs_rhs_info(m, n, 2, 4, 4, 1, 2, true, 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);
-                }
+                return configure_lhs_rhs_info(m, n, 4, 2, 4, 2, 2, false, false, true, false, false);
             }
         }
         else
         {
-            if(r_mn <= 0.21f)
+            if(r_nk <= 0.67f)
             {
-                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);
-                }
+                return configure_lhs_rhs_info(m, n, 4, 2, 4, 2, 2, false, false, true, false, false);
             }
             else
             {
-                return configure_lhs_rhs_info(m, n, 2, 4, 4, 2, 2, false, false, true, false, false);
+                return configure_lhs_rhs_info(m, n, 2, 4, 4, 4, 1, false, true, false, true, false);
             }
         }
     }
     else
     {
-        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);
-                }
-            }
-        }
+        return configure_lhs_rhs_info(m, n, 4, 4, 4, 2, 4, true, true, true, false, false);
     }
 }
 
-- 
cgit v1.2.1