From 229757be70e7884b1a64972c2bdf9919075401a2 Mon Sep 17 00:00:00 2001
From: Gian Marco Iodice <gianmarco.iodice@arm.com>
Date: Sun, 15 Nov 2020 16:06:10 +0000
Subject: COMPMID-3975: Update CLGEMM heuristic for Mali-G52 F16

- Update the GEMM heuristic for Mali-G52
- Fix the GEMV heuristic on Mali-G76

Change-Id: I895153d02d09e73c51ecf158e2ae43f81f6d7b05
Signed-off-by: Gian Marco Iodice <gianmarco.iodice@arm.com>
Reviewed-on: https://review.mlplatform.org/c/ml/ComputeLibrary/+/4425
Tested-by: Arm Jenkins <bsgcomp@arm.com>
Reviewed-by: Michele Di Giorgio <michele.digiorgio@arm.com>
Comments-Addressed: Arm Jenkins <bsgcomp@arm.com>
---
 .../CL/gemm/CLGEMMKernelSelectionBifrost.cpp       | 122 ++++++++++++++++++++-
 1 file changed, 121 insertions(+), 1 deletion(-)

(limited to 'src/runtime/CL/gemm/CLGEMMKernelSelectionBifrost.cpp')

diff --git a/src/runtime/CL/gemm/CLGEMMKernelSelectionBifrost.cpp b/src/runtime/CL/gemm/CLGEMMKernelSelectionBifrost.cpp
index c77746a044..46d07fffba 100644
--- a/src/runtime/CL/gemm/CLGEMMKernelSelectionBifrost.cpp
+++ b/src/runtime/CL/gemm/CLGEMMKernelSelectionBifrost.cpp
@@ -72,7 +72,7 @@ CLGEMMKernelType CLGEMMKernelSelectionBifrost::select_kernel(const CLGEMMKernelS
     static std::map<DataType, FunctionExecutorPtr> gemm_g52_configs =
     {
         { DataType::F32, &CLGEMMKernelSelectionBifrost::g52_f32 },
-        { DataType::F16, &CLGEMMKernelSelectionBifrost::default_f16 },
+        { DataType::F16, &CLGEMMKernelSelectionBifrost::g52_f16 },
         { DataType::QASYMM8, &CLGEMMKernelSelectionBifrost::default_q8 },
         { DataType::QASYMM8_SIGNED, &CLGEMMKernelSelectionBifrost::default_q8 },
         { DataType::QSYMM8, &CLGEMMKernelSelectionBifrost::default_q8 },
@@ -443,6 +443,126 @@ CLGEMMKernelType CLGEMMKernelSelectionBifrost::g76_f16(unsigned int m, unsigned
     }
 }
 
+CLGEMMKernelType CLGEMMKernelSelectionBifrost::g52_f16(unsigned int m, unsigned int n, unsigned int k, unsigned int b, bool is_rhs_constant)
+{
+    ARM_COMPUTE_UNUSED(b);
+
+    if (!is_rhs_constant)
+    {
+        return CLGEMMKernelType::NATIVE_V1;
+    }
+
+    if (m == 1)
+    {
+        return CLGEMMKernelType::RESHAPED_ONLY_RHS;
+    }
+
+    const float r_mn  = static_cast<float>(m) / static_cast<float>(n);
+    const float r_mk  = static_cast<float>(m) / static_cast<float>(k);
+    const float r_nk  = static_cast<float>(n) / static_cast<float>(k);
+    const float r_mnk = static_cast<float>(m) / (static_cast<float>(n) * static_cast<float>(k));
+
+    if(r_mn <= 22.9200f)
+    {
+        if(r_mk <= 0.0157f)
+        {
+            return CLGEMMKernelType::RESHAPED_ONLY_RHS;
+        }
+        else
+        {
+            if(r_mnk <= 7809.3750f)
+            {
+                if(r_mnk <= 101.7937f)
+                {
+                    if(r_mn <= 0.4594f)
+                    {
+                        if(r_mk <= 0.0557f)
+                        {
+                            return CLGEMMKernelType::RESHAPED_ONLY_RHS;
+                        }
+                        else
+                        {
+                            return CLGEMMKernelType::RESHAPED;
+                        }
+                    }
+                    else
+                    {
+                        return CLGEMMKernelType::RESHAPED_ONLY_RHS;
+                    }
+                }
+                else
+                {
+                    if(r_nk <= 0.4396f)
+                    {
+                        if(r_mn <= 1.5182f)
+                        {
+                            if(r_mnk <= 1709.9167f)
+                            {
+                                return CLGEMMKernelType::RESHAPED;
+                            }
+                            else
+                            {
+                                return CLGEMMKernelType::RESHAPED_ONLY_RHS;
+                            }
+                        }
+                        else
+                        {
+                            if(r_mnk <= 1330.6000f)
+                            {
+                                return CLGEMMKernelType::RESHAPED_ONLY_RHS;
+                            }
+                            else
+                            {
+                                return CLGEMMKernelType::RESHAPED;
+                            }
+                        }
+                    }
+                    else
+                    {
+                        if(r_mn <= 2.5896f)
+                        {
+                            return CLGEMMKernelType::RESHAPED;
+                        }
+                        else
+                        {
+                            if(r_mnk <= 326.6667f)
+                            {
+                                return CLGEMMKernelType::RESHAPED_ONLY_RHS;
+                            }
+                            else
+                            {
+                                return CLGEMMKernelType::RESHAPED;
+                            }
+                        }
+                    }
+                }
+            }
+            else
+            {
+                return CLGEMMKernelType::RESHAPED_ONLY_RHS;
+            }
+        }
+    }
+    else
+    {
+        if(r_mn <= 86.7578f)
+        {
+            if(r_mnk <= 11231.6406f)
+            {
+                return CLGEMMKernelType::RESHAPED_ONLY_RHS;
+            }
+            else
+            {
+                return CLGEMMKernelType::RESHAPED;
+            }
+        }
+        else
+        {
+            return CLGEMMKernelType::RESHAPED_ONLY_RHS;
+        }
+    }
+}
+
 CLGEMMKernelType CLGEMMKernelSelectionBifrost::g71_f16(unsigned int m, unsigned int n, unsigned int k, unsigned int b, bool is_rhs_constant)
 {
     ARM_COMPUTE_UNUSED(b);
-- 
cgit v1.2.1