commit 839e19865d4b654899d1da5cfb94304841e7f210
author Gian Marco Iodice <gianmarco.iodice@arm.com> Thu Oct 29 13:36:50 2020 +0000
committer Gian Marco Iodice <gianmarco.iodice@arm.com> Fri Oct 30 15:35:02 2020 +0000
tree 10321574df9e263036a60689fb5fb03608b2f487
parent c4d45559b00cdbdca80296c23be5939439fbbbd0

COMPMID-3930: Update CLGEMM heuristic for fp16. Mali-G76

- Since the GEMM kernel can now work without padding, the heuristic
  requires to be fine-tuned to exploit this feature
- The heuristic affects Mali-G76 FP16 only

Change-Id: Ia430627f02131ad956ce2219b80c83c8e7cabaf2
Signed-off-by: Gian Marco Iodice <gianmarco.iodice@arm.com>
Reviewed-on: https://review.mlplatform.org/c/ml/ComputeLibrary/+/4284
Tested-by: Arm Jenkins <bsgcomp@arm.com>
Comments-Addressed: Arm Jenkins <bsgcomp@arm.com>
Reviewed-by: Michele Di Giorgio <michele.digiorgio@arm.com>
Reviewed-by: SiCong Li <sicong.li@arm.com>

src/core/CL/gemm/CLGEMMHelpers.cpp

diff --git a/src/core/CL/gemm/CLGEMMHelpers.cpp b/src/core/CL/gemm/CLGEMMHelpers.cpp
index 877bf1e..d60626b 100644
--- a/src/core/CL/gemm/CLGEMMHelpers.cpp
+++ b/src/core/CL/gemm/CLGEMMHelpers.cpp
@@ -27,6 +27,7 @@
 #include "arm_compute/core/CL/CLKernelLibrary.h"
 #include "arm_compute/core/CL/OpenCL.h"
 #include "arm_compute/core/ITensorInfo.h"
+#include "arm_compute/core/utils/misc/ShapeCalculator.h"
 
 #include <utility>
 
@@ -34,11 +35,13 @@
 {
 namespace cl_gemm
 {
+using namespace arm_compute::misc::shape_calculator;
+
 std::pair<GEMMLHSMatrixInfo, GEMMRHSMatrixInfo> configure_lhs_rhs_info(unsigned int m, unsigned int n, unsigned int m0, unsigned int n0, unsigned int k0, unsigned int v0, unsigned int h0,
                                                                        bool lhs_interleave, bool rhs_interleave, bool lhs_transpose, bool rhs_transpose, bool export_to_cl_image)
 {
-    v0 = ((m / (m0 * v0)) == 0) ? 1 : v0;
-    h0 = ((n / (n0 * h0)) == 0) ? 1 : h0;
+    v0 = std::max(std::min(static_cast<int>(m / m0), static_cast<int>(v0)), static_cast<int>(1));
+    h0 = std::max(std::min(static_cast<int>(n / n0), static_cast<int>(h0)), static_cast<int>(1));
 
     const GEMMLHSMatrixInfo lhs_info(m0, k0, v0, lhs_transpose, lhs_interleave);
     const GEMMRHSMatrixInfo rhs_info(n0, k0, h0, rhs_transpose, rhs_interleave, export_to_cl_image);
@@ -46,6 +49,24 @@
     return std::make_pair(lhs_info, rhs_info);
 }
 
+std::pair<GEMMLHSMatrixInfo, GEMMRHSMatrixInfo> select_lhs_rhs_info(std::pair<GEMMLHSMatrixInfo, GEMMRHSMatrixInfo> info_img,
+                                                                    std::pair<GEMMLHSMatrixInfo, GEMMRHSMatrixInfo> info_buf,
+                                                                    unsigned int n, unsigned int k, unsigned int b, DataType data_type)
+{
+    const TensorInfo  tensor_rhs_info(TensorShape(n, k, b), 1, data_type);
+    const TensorShape shape = compute_rhs_reshaped_shape(tensor_rhs_info, info_img.second);
+    const TensorInfo  tensor_reshaped_info(shape, 1, data_type);
+
+    if(bool(validate_image2d_support_on_rhs(tensor_reshaped_info, info_img.second)))
+    {
+        return info_img;
+    }
+    else
+    {
+        return info_buf;
+    }
+}
+
 void update_padding_for_cl_image(ITensorInfo *tensor)
 {
     constexpr unsigned int num_floats_per_pixel = 4;