commit ee6454a3dfc5d9e921689f0b60cfa15ad4ef6654
author Gian Marco Iodice <gianmarco.iodice@arm.com> Tue Sep 17 10:56:51 2019 +0100
committer Gian Marco Iodice <gianmarco.iodice@arm.com> Mon Sep 23 13:39:11 2019 +0000
tree 9d534390e15b5b22112735a3572c4df10be8a275
parent 4c268b97dff93eae3f71f2a6971f0d3f748b7b38

COMPMID-2668: Add GEMM/GEMMLowp heuristic for Mali-G51

Change-Id: I920ee5ffeff86806b9ac0620250dba98d1d081ba
Signed-off-by: Gian Marco Iodice <gianmarco.iodice@arm.com>
Reviewed-on: https://review.mlplatform.org/c/1955
Reviewed-by: Georgios Pinitas <georgios.pinitas@arm.com>
Comments-Addressed: Arm Jenkins <bsgcomp@arm.com>
Tested-by: Arm Jenkins <bsgcomp@arm.com>

src/core/CL/gemm/reshaped_only_rhs/CLGEMMReshapedOnlyRHSKernelConfigurationBifrost.cpp

diff --git a/src/core/CL/gemm/reshaped_only_rhs/CLGEMMReshapedOnlyRHSKernelConfigurationBifrost.cpp b/src/core/CL/gemm/reshaped_only_rhs/CLGEMMReshapedOnlyRHSKernelConfigurationBifrost.cpp
index 483bab8..9f3fc3a 100644
--- a/src/core/CL/gemm/reshaped_only_rhs/CLGEMMReshapedOnlyRHSKernelConfigurationBifrost.cpp
+++ b/src/core/CL/gemm/reshaped_only_rhs/CLGEMMReshapedOnlyRHSKernelConfigurationBifrost.cpp
@@ -48,6 +48,13 @@
     using ConfigurationFunctionExecutorPtr = std::pair<GEMMLHSMatrixInfo, GEMMRHSMatrixInfo> (CLGEMMReshapedOnlyRHSKernelConfigurationBifrost::*)(unsigned int m, unsigned int n, unsigned int k,
                                              unsigned int b);
 
+    // Configurations for Mali-G51
+    static std::map<DataType, ConfigurationFunctionExecutorPtr> gemm_configs_G51 =
+    {
+        { DataType::F32, &CLGEMMReshapedOnlyRHSKernelConfigurationBifrost::configure_G51_f32 },
+        { DataType::QASYMM8, &CLGEMMReshapedOnlyRHSKernelConfigurationBifrost::configure_G51_u8 }
+    };
+
     // Configurations for Mali-G76
     static std::map<DataType, ConfigurationFunctionExecutorPtr> gemm_configs_G76 =
     {
@@ -66,6 +73,8 @@
     {
         case GPUTarget::G76:
             return (this->*gemm_configs_G76[data_type])(m, n, k, b);
+        case GPUTarget::G51:
+            return (this->*gemm_configs_G51[data_type])(m, n, k, b);
         default:
             return (this->*gemm_configs_G7x[data_type])(m, n, k, b);
     }
@@ -111,6 +120,23 @@
     }
 }
 
+std::pair<GEMMLHSMatrixInfo, GEMMRHSMatrixInfo> CLGEMMReshapedOnlyRHSKernelConfigurationBifrost::configure_G51_f32(unsigned int m, unsigned int n, unsigned int k, unsigned int b)
+{
+    ARM_COMPUTE_UNUSED(k);
+    ARM_COMPUTE_UNUSED(b);
+
+    if(m == 1)
+    {
+        const unsigned int n0 = n < 1280? 2 : 4;
+        const unsigned int h0 = std::max(n / n0, static_cast<unsigned int>(1));
+        return configure_lhs_rhs_info(m, n, 1, n0, 4, 1, h0, false, true, false, true);
+    }
+    else
+    {
+        return configure_lhs_rhs_info(m, n, 4, 4, 4, 1, 2, false, true, false, true);
+    }
+}
+
 std::pair<GEMMLHSMatrixInfo, GEMMRHSMatrixInfo> CLGEMMReshapedOnlyRHSKernelConfigurationBifrost::configure_G7x_u8(unsigned int m, unsigned int n, unsigned int k, unsigned int b)
 {
     ARM_COMPUTE_UNUSED(k);
@@ -159,5 +185,22 @@
         return configure_lhs_rhs_info(m, n, 4, 4, 16, 1, 2, false, true, false, true);
     }
 }
+
+std::pair<GEMMLHSMatrixInfo, GEMMRHSMatrixInfo> CLGEMMReshapedOnlyRHSKernelConfigurationBifrost::configure_G51_u8(unsigned int m, unsigned int n, unsigned int k, unsigned int b)
+{
+    ARM_COMPUTE_UNUSED(k);
+    ARM_COMPUTE_UNUSED(b);
+
+    if(m == 1)
+    {
+        const unsigned int h0 = std::max(n / 2, static_cast<unsigned int>(1));
+        return configure_lhs_rhs_info(m, n, 1, 4, 16, 1, h0, false, true, false, true);
+    }
+    else
+    {
+        const unsigned int h0 = std::max(n / 2, static_cast<unsigned int>(1));
+        return configure_lhs_rhs_info(m, n, 4, 2, 16, 1, h0, false, true, false, true);
+    }
+}
 } // namespace cl_gemm
 } // namespace arm_compute
\ No newline at end of file