From eb65f6da695ac0d3e495817145cceb1c4de4f048 Mon Sep 17 00:00:00 2001
From: Gian Marco Iodice <gianmarco.iodice@arm.com>
Date: Wed, 15 Apr 2020 11:42:15 +0100
Subject: COMPMID-3304: Update OpenCL GEMM heuristic for Int8

Change-Id: I6b7ff678d8d0437a1639db2ff602ea1cdb155464
Signed-off-by: Gian Marco Iodice <gianmarco.iodice@arm.com>
Reviewed-on: https://review.mlplatform.org/c/ml/ComputeLibrary/+/3056
Tested-by: Arm Jenkins <bsgcomp@arm.com>
Reviewed-by: Georgios Pinitas <georgios.pinitas@arm.com>
Comments-Addressed: Arm Jenkins <bsgcomp@arm.com>
---
 ...GEMMReshapedOnlyRHSKernelConfigurationBifrost.cpp | 20 ++++++++++++++------
 1 file changed, 14 insertions(+), 6 deletions(-)

(limited to '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 8e798116bf..8826cca11b 100644
--- a/src/core/CL/gemm/reshaped_only_rhs/CLGEMMReshapedOnlyRHSKernelConfigurationBifrost.cpp
+++ b/src/core/CL/gemm/reshaped_only_rhs/CLGEMMReshapedOnlyRHSKernelConfigurationBifrost.cpp
@@ -50,7 +50,10 @@ std::pair<GEMMLHSMatrixInfo, GEMMRHSMatrixInfo> CLGEMMReshapedOnlyRHSKernelConfi
     {
         { DataType::F32, &CLGEMMReshapedOnlyRHSKernelConfigurationBifrost::configure_G51_f32 },
         { DataType::F16, &CLGEMMReshapedOnlyRHSKernelConfigurationBifrost::configure_G51_f16 },
-        { DataType::QASYMM8, &CLGEMMReshapedOnlyRHSKernelConfigurationBifrost::configure_G51_u8 }
+        { DataType::QASYMM8, &CLGEMMReshapedOnlyRHSKernelConfigurationBifrost::configure_G51_u8 },
+        { DataType::QSYMM8, &CLGEMMReshapedOnlyRHSKernelConfigurationBifrost::configure_G51_u8 },
+        { DataType::QASYMM8_SIGNED, &CLGEMMReshapedOnlyRHSKernelConfigurationBifrost::configure_G51_u8 },
+        { DataType::QSYMM8_PER_CHANNEL, &CLGEMMReshapedOnlyRHSKernelConfigurationBifrost::configure_G51_u8 }
     };
 
     // Configurations for Mali-G76
@@ -58,7 +61,10 @@ std::pair<GEMMLHSMatrixInfo, GEMMRHSMatrixInfo> CLGEMMReshapedOnlyRHSKernelConfi
     {
         { DataType::F32, &CLGEMMReshapedOnlyRHSKernelConfigurationBifrost::configure_G76_f32 },
         { DataType::F16, &CLGEMMReshapedOnlyRHSKernelConfigurationBifrost::configure_G76_f16 },
-        { DataType::QASYMM8, &CLGEMMReshapedOnlyRHSKernelConfigurationBifrost::configure_G76_u8 }
+        { DataType::QASYMM8, &CLGEMMReshapedOnlyRHSKernelConfigurationBifrost::configure_G76_u8 },
+        { DataType::QSYMM8, &CLGEMMReshapedOnlyRHSKernelConfigurationBifrost::configure_G76_u8 },
+        { DataType::QASYMM8_SIGNED, &CLGEMMReshapedOnlyRHSKernelConfigurationBifrost::configure_G76_u8 },
+        { DataType::QSYMM8_PER_CHANNEL, &CLGEMMReshapedOnlyRHSKernelConfigurationBifrost::configure_G76_u8 }
     };
 
     // Configurations for Mali-G7x
@@ -66,7 +72,10 @@ std::pair<GEMMLHSMatrixInfo, GEMMRHSMatrixInfo> CLGEMMReshapedOnlyRHSKernelConfi
     {
         { DataType::F32, &CLGEMMReshapedOnlyRHSKernelConfigurationBifrost::configure_G7x_f32 },
         { DataType::F16, &CLGEMMReshapedOnlyRHSKernelConfigurationBifrost::configure_G7x_f16 },
-        { DataType::QASYMM8, &CLGEMMReshapedOnlyRHSKernelConfigurationBifrost::configure_G7x_u8 }
+        { DataType::QASYMM8, &CLGEMMReshapedOnlyRHSKernelConfigurationBifrost::configure_G7x_u8 },
+        { DataType::QSYMM8, &CLGEMMReshapedOnlyRHSKernelConfigurationBifrost::configure_G7x_u8 },
+        { DataType::QASYMM8_SIGNED, &CLGEMMReshapedOnlyRHSKernelConfigurationBifrost::configure_G7x_u8 },
+        { DataType::QSYMM8_PER_CHANNEL, &CLGEMMReshapedOnlyRHSKernelConfigurationBifrost::configure_G7x_u8 }
     };
 
     switch(_target)
@@ -235,15 +244,14 @@ std::pair<GEMMLHSMatrixInfo, GEMMRHSMatrixInfo> CLGEMMReshapedOnlyRHSKernelConfi
     }
     else
     {
+        const int h0 = std::max(std::min(static_cast<int>(n / 2), static_cast<int>(128)), static_cast<int>(1));
         if(m == 1)
         {
-            const unsigned int h0 = std::max(n / 2, 1U);
             return configure_lhs_rhs_info(m, n, 1, 2, 4, 1, h0, false, true, false, true);
         }
         else
         {
-            const unsigned int h0 = std::max(n / 4, 1U);
-            return configure_lhs_rhs_info(m, n, 2, 2, 16, 1, h0, false, true, false, true);
+            return configure_lhs_rhs_info(m, n, 4, 2, 16, 1, h0, false, true, false, true);
         }
     }
 }
-- 
cgit v1.2.1