COMPMID-4018: Fix heuristic fallback for CLGEMMReshapedRHSOnly for
Mali-G52
- Missing fallback in case of export to cl_image
Change-Id: I5bb3013fd1350628f16e4709c4bb31999fece22d
Signed-off-by: Gian Marco Iodice <gianmarco.iodice@arm.com>
Reviewed-on: https://review.mlplatform.org/c/ml/ComputeLibrary/+/4531
Reviewed-by: Pablo Marquez Tello <pablo.tello@arm.com>
Reviewed-by: Michele Di Giorgio <michele.digiorgio@arm.com>
Tested-by: Arm Jenkins <bsgcomp@arm.com>
Comments-Addressed: Arm Jenkins <bsgcomp@arm.com>
diff --git a/src/core/CL/gemm/reshaped_only_rhs/CLGEMMReshapedOnlyRHSKernelConfigurationBifrost.cpp b/src/core/CL/gemm/reshaped_only_rhs/CLGEMMReshapedOnlyRHSKernelConfigurationBifrost.cpp
index a2c1ed2..d5b76d8 100644
--- a/src/core/CL/gemm/reshaped_only_rhs/CLGEMMReshapedOnlyRHSKernelConfigurationBifrost.cpp
+++ b/src/core/CL/gemm/reshaped_only_rhs/CLGEMMReshapedOnlyRHSKernelConfigurationBifrost.cpp
@@ -322,8 +322,15 @@
const float r_mk = static_cast<float>(m) / static_cast<float>(k);
const float r_nk = static_cast<float>(n) / static_cast<float>(k);
+ GEMMLHSMatrixInfo lhs_info_buf;
+ GEMMRHSMatrixInfo rhs_info_buf;
+ GEMMLHSMatrixInfo lhs_info_img;
+ GEMMRHSMatrixInfo rhs_info_img;
+
if(m == 1)
{
+ std::tie(lhs_info_buf, rhs_info_buf) = configure_lhs_rhs_info(m, n, 1, 4, 16, 1, 16, false, true, false, false, false);
+
if(r_mk <= 0.0026f)
{
if(r_nk <= 0.4664f)
@@ -332,7 +339,10 @@
}
else
{
- return configure_lhs_rhs_info(m, n, 1, 4, 16, 1, 16, false, true, false, false, true);
+ std::tie(lhs_info_img, rhs_info_img) = configure_lhs_rhs_info(m, n, 1, 4, 16, 1, 16, false, true, false, false, true);
+ return select_lhs_rhs_info(std::make_pair(lhs_info_img, rhs_info_img),
+ std::make_pair(lhs_info_buf, rhs_info_buf),
+ n, k, b, DataType::F16);
}
}
else
@@ -343,12 +353,17 @@
}
else
{
- return configure_lhs_rhs_info(m, n, 1, 4, 16, 1, 16, false, true, false, false, true);
+ std::tie(lhs_info_img, rhs_info_img) = configure_lhs_rhs_info(m, n, 1, 4, 16, 1, 16, false, true, false, false, true);
+ return select_lhs_rhs_info(std::make_pair(lhs_info_img, rhs_info_img),
+ std::make_pair(lhs_info_buf, rhs_info_buf),
+ n, k, b, DataType::F16);
}
}
}
else
{
+ std::tie(lhs_info_buf, rhs_info_buf) = configure_lhs_rhs_info(m, n, 5, 8, 4, 1, 2, false, false, false, false, false);
+
if(workload <= 362.6000f)
{
return configure_lhs_rhs_info(m, n, 2, 2, 8, 1, 16, false, false, false, true, false);
@@ -359,7 +374,10 @@
{
if(workload <= 708.8000f)
{
- return configure_lhs_rhs_info(m, n, 5, 4, 4, 1, 2, false, false, false, false, true);
+ std::tie(lhs_info_img, rhs_info_img) = configure_lhs_rhs_info(m, n, 5, 4, 4, 1, 2, false, false, false, false, true);
+ return select_lhs_rhs_info(std::make_pair(lhs_info_img, rhs_info_img),
+ std::make_pair(lhs_info_buf, rhs_info_buf),
+ n, k, b, DataType::F16);
}
else
{
@@ -374,7 +392,10 @@
}
else
{
- return configure_lhs_rhs_info(m, n, 5, 4, 4, 1, 2, false, false, false, false, true);
+ std::tie(lhs_info_img, rhs_info_img) = configure_lhs_rhs_info(m, n, 5, 4, 4, 1, 2, false, false, false, false, true);
+ return select_lhs_rhs_info(std::make_pair(lhs_info_img, rhs_info_img),
+ std::make_pair(lhs_info_buf, rhs_info_buf),
+ n, k, b, DataType::F16);
}
}
}