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>

1 files changed, 130 insertions, 6 deletions

diff --git a/src/runtime/CL/gemm/CLGEMMKernelSelectionValhall.cpp b/src/runtime/CL/gemm/CLGEMMKernelSelectionValhall.cpp
index acae0e7565..da41859b87 100644
--- a/src/runtime/CL/gemm/CLGEMMKernelSelectionValhall.cpp
+++ b/src/runtime/CL/gemm/CLGEMMKernelSelectionValhall.cpp
@@ -46,8 +46,8 @@ CLGEMMKernelType CLGEMMKernelSelectionValhall::select_kernel(const CLGEMMKernelS
 
     using FunctionExecutorPtr = CLGEMMKernelType (CLGEMMKernelSelectionValhall::*)(unsigned int m, unsigned int n, unsigned int k, unsigned int b, bool is_rhs_constant);
 
-    // Configurations for Valhall architectures
-    static std::map<DataType, FunctionExecutorPtr> gemm_configs =
+    // Default configurations for Valhall architectures
+    static std::map<DataType, FunctionExecutorPtr> gemm_default_configs =
     {
         { DataType::F32, &CLGEMMKernelSelectionValhall::default_f32 },
         { DataType::F16, &CLGEMMKernelSelectionValhall::default_f16 },
@@ -57,14 +57,34 @@ CLGEMMKernelType CLGEMMKernelSelectionValhall::select_kernel(const CLGEMMKernelS
         { DataType::QSYMM8_PER_CHANNEL, &CLGEMMKernelSelectionValhall::default_q8 }
     };
 
+    // Mali-G77 configurations
+    static std::map<DataType, FunctionExecutorPtr> gemm_g77_configs =
+    {
+        { DataType::F32, &CLGEMMKernelSelectionValhall::default_f32 },
+        { DataType::F16, &CLGEMMKernelSelectionValhall::g77_f16 },
+        { DataType::QASYMM8, &CLGEMMKernelSelectionValhall::default_q8 },
+        { DataType::QASYMM8_SIGNED, &CLGEMMKernelSelectionValhall::default_q8 },
+        { DataType::QSYMM8, &CLGEMMKernelSelectionValhall::default_q8 },
+        { DataType::QSYMM8_PER_CHANNEL, &CLGEMMKernelSelectionValhall::default_q8 }
+    };
+
     const DataType data_type = params.data_type;
 
-    if(gemm_configs.find(data_type) != gemm_configs.end())
+    switch(_target)
     {
-        return (this->*gemm_configs[data_type])(params.m, params.n, params.k, params.b, params.is_rhs_constant);
+        case GPUTarget::G77:
+            if(gemm_g77_configs.find(data_type) != gemm_g77_configs.end())
+            {
+                return (this->*gemm_g77_configs[data_type])(params.m, params.n, params.k, params.b, params.is_rhs_constant);
+            }
+            ARM_COMPUTE_ERROR("Not supported data type");
+        default:
+            if(gemm_default_configs.find(data_type) != gemm_default_configs.end())
+            {
+                return (this->*gemm_default_configs[data_type])(params.m, params.n, params.k, params.b, params.is_rhs_constant);
+            }
+            ARM_COMPUTE_ERROR("Not supported data type");
     }
-
-    ARM_COMPUTE_ERROR("Not supported data type");
 }
 
 CLGEMMKernelType CLGEMMKernelSelectionValhall::default_f32(unsigned int m, unsigned int n, unsigned int k, unsigned int b, bool is_rhs_constant)
@@ -81,6 +101,110 @@ CLGEMMKernelType CLGEMMKernelSelectionValhall::default_f16(unsigned int m, unsig
     return is_rhs_constant ? CLGEMMKernelType::RESHAPED_ONLY_RHS : CLGEMMKernelType::NATIVE_V1;
 }
 
+CLGEMMKernelType CLGEMMKernelSelectionValhall::g77_f16(unsigned int m, unsigned int n, unsigned int k, unsigned int b, bool is_rhs_constant)
+{
+    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 workload = (static_cast<float>(m) * static_cast<float>(n) * static_cast<float>(b)) / 20.0f;
+
+    if(r_mk <= 0.6817956566810608)
+    {
+        if(workload <= 801.6000061035156)
+        {
+            return CLGEMMKernelType::RESHAPED_ONLY_RHS;
+        }
+        else
+        {
+            if(r_mn <= 0.0839829258620739)
+            {
+                return CLGEMMKernelType::RESHAPED_ONLY_RHS;
+            }
+            else
+            {
+                if(r_mk <= 0.24917218834161758)
+                {
+                    return CLGEMMKernelType::RESHAPED;
+                }
+                else
+                {
+                    if(workload <= 2551.75)
+                    {
+                        return CLGEMMKernelType::RESHAPED_ONLY_RHS;
+                    }
+                    else
+                    {
+                        if(workload <= 5061.574951171875)
+                        {
+                            return CLGEMMKernelType::RESHAPED_ONLY_RHS;
+                        }
+                        else
+                        {
+                            return CLGEMMKernelType::RESHAPED;
+                        }
+                    }
+                }
+            }
+        }
+    }
+    else
+    {
+        if(r_mk <= 4.849947690963745)
+        {
+            if(workload <= 17618.4501953125)
+            {
+                if(workload <= 5224.699951171875)
+                {
+                    return CLGEMMKernelType::RESHAPED_ONLY_RHS;
+                }
+                else
+                {
+                    if(r_nk <= 0.7933054566383362)
+                    {
+                        return CLGEMMKernelType::RESHAPED;
+                    }
+                    else
+                    {
+                        return CLGEMMKernelType::RESHAPED_ONLY_RHS;
+                    }
+                }
+            }
+            else
+            {
+                if(workload <= 20275.2001953125)
+                {
+                    return CLGEMMKernelType::RESHAPED;
+                }
+                else
+                {
+                    if(r_mk <= 3.07421875)
+                    {
+                        return CLGEMMKernelType::RESHAPED_ONLY_RHS;
+                    }
+                    else
+                    {
+                        return CLGEMMKernelType::RESHAPED;
+                    }
+                }
+            }
+        }
+        else
+        {
+            return CLGEMMKernelType::RESHAPED_ONLY_RHS;
+        }
+    }
+}
+
 CLGEMMKernelType CLGEMMKernelSelectionValhall::default_q8(unsigned int m, unsigned int n, unsigned int k, unsigned int b, bool is_rhs_constant)
 {
     ARM_COMPUTE_UNUSED(m, n, k, b);