From f1f61444429447e207e6f145b9d9645d0216dede Mon Sep 17 00:00:00 2001 From: Gian Marco Iodice Date: Wed, 18 Nov 2020 13:35:21 +0000 Subject: COMPMID-3980: Fix F16 regression on Mali-G76 - The issue was related to the wrong N0 used in CLGEMMReshaped heuristic Change-Id: Iecb010f0de383169c9fb8543faf15b2f93ba2ed8 Signed-off-by: Gian Marco Iodice Reviewed-on: https://review.mlplatform.org/c/ml/ComputeLibrary/+/4464 Reviewed-by: Georgios Pinitas Comments-Addressed: Arm Jenkins Tested-by: Arm Jenkins --- .../CLGEMMReshapedKernelConfigurationBifrost.cpp | 18 +++++------------- 1 file changed, 5 insertions(+), 13 deletions(-) diff --git a/src/core/CL/gemm/reshaped/CLGEMMReshapedKernelConfigurationBifrost.cpp b/src/core/CL/gemm/reshaped/CLGEMMReshapedKernelConfigurationBifrost.cpp index e5c89f89af..59a2a82edf 100644 --- a/src/core/CL/gemm/reshaped/CLGEMMReshapedKernelConfigurationBifrost.cpp +++ b/src/core/CL/gemm/reshaped/CLGEMMReshapedKernelConfigurationBifrost.cpp @@ -333,13 +333,12 @@ std::pair CLGEMMReshapedKernelConfiguratio { const float workload = (static_cast(m) * static_cast(n) * static_cast(b)) / 20.0f; const float r_mk = static_cast(m) / static_cast(k); - const float r_nk = static_cast(n) / static_cast(k); - if(workload <= 1422.40f) + if(workload <= 1595.2000f) { - if(r_mk <= 2.45f) + if(r_mk <= 2.1044f) { - if(workload <= 801.60f) + if(workload <= 870.4000f) { return configure_lhs_rhs_info(m, n, 2, 4, 4, 1, 2, true, false, true, false, false); } @@ -350,19 +349,12 @@ std::pair CLGEMMReshapedKernelConfiguratio } else { - if(r_nk <= 0.67f) - { - 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, 4, 1, false, true, false, true, 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, 4, 4, 4, 2, 4, true, true, true, false, false); + return configure_lhs_rhs_info(m, n, 4, 8, 4, 4, 2, true, true, true, false, false); } } -- cgit v1.2.1