COMPMID-809: Add NHWC data format on CLGEMMConvolutionLayer.

Change-Id: I50e4f5e7d47e21c300f754bee2c216863075b5cf
Reviewed-on: https://eu-gerrit-1.euhpc.arm.com/136191
Tested-by: Jenkins <bsgcomp@arm.com>
Reviewed-by: Giorgio Arena <giorgio.arena@arm.com>
Reviewed-by: Gian Marco Iodice <gianmarco.iodice@arm.com>

3 files changed, 31 insertions, 11 deletions

diff --git a/arm_compute/core/TensorShape.h b/arm_compute/core/TensorShape.h
index 0c3d9414e1..0340e1a644 100644
--- a/arm_compute/core/TensorShape.h
+++ b/arm_compute/core/TensorShape.h
@@ -136,6 +136,20 @@ public:
         // Make sure all empty dimensions are filled with 1
         std::fill(_id.begin() + _num_dimensions, _id.end(), 1);
     }
+    /** Shifts right the tensor shape increasing its dimensions
+     *
+     * @param[in] step Rotation step
+     */
+    void shift_right(size_t step)
+    {
+        ARM_COMPUTE_ERROR_ON(step > TensorShape::num_max_dimensions - num_dimensions());
+
+        std::rotate(begin(), begin() + TensorShape::num_max_dimensions - step, end());
+        _num_dimensions += step;
+
+        // Correct number dimensions to ignore trailing dimensions of size 1
+        apply_dimension_correction();
+    }
 
     /** Return a copy with collapsed dimensions starting from a given point.
      *
diff --git a/arm_compute/core/utils/misc/ShapeCalculator.h b/arm_compute/core/utils/misc/ShapeCalculator.h
index f64cf9d6ae..115cbe688d 100644
--- a/arm_compute/core/utils/misc/ShapeCalculator.h
+++ b/arm_compute/core/utils/misc/ShapeCalculator.h
@@ -110,6 +110,7 @@ inline TensorShape compute_reductionB_shape(const ITensorInfo &a)
 inline TensorShape compute_col2im_shape(const ITensorInfo &input, std::pair<unsigned int, unsigned int> convolved_dims)
 {
     TensorShape col2im_shape{ input.tensor_shape() };
+    col2im_shape.shift_right(1);
     col2im_shape.set(0, convolved_dims.first);
     col2im_shape.set(1, convolved_dims.second);
     col2im_shape.set(2, input.tensor_shape()[0]);
diff --git a/arm_compute/runtime/CL/functions/CLGEMMConvolutionLayer.h b/arm_compute/runtime/CL/functions/CLGEMMConvolutionLayer.h
index 3dde52989b..2c1f7a9d5e 100644
--- a/arm_compute/runtime/CL/functions/CLGEMMConvolutionLayer.h
+++ b/arm_compute/runtime/CL/functions/CLGEMMConvolutionLayer.h
@@ -158,22 +158,24 @@ public:
 private:
     /** Configures the appropriate matrix multiply routine
      *
-     * @param input                     Input tensor. Data types supported: QS8/QASYMM8/QS16/F16/F32.
-     * @param weights                   Weights tensor. Data type supported: Same as @p input.
-     * @param output                    Output tensor. Data types supported: Same as @p input,
-     *                                                 except for input of QASYMM8 type where output should be of S32 type.
+     * @param[in]      input         Input tensor. Data types supported: QS8/QASYMM8/QS16/F16/F32.
+     * @param[in]      weights       Weights tensor. Data type supported: Same as @p input.
+     * @param[in, out] output        Output tensor. Data types supported: Same as @p input,
+     *                               except for input of QASYMM8 type where output should be of S32 type.
+     * @param[in]      gemm_3d_depth (Optional) Depth of GEMM 3D (Defaults to 1)
      */
-    void configure_mm(const ICLTensor *input, const ICLTensor *weights, ICLTensor *output);
+    void configure_mm(const ICLTensor *input, const ICLTensor *weights, ICLTensor *output, int gemm_3d_depth = 1);
     /** Static function to check if given info will lead to a valid configuration of @ref CLGEMMConvolutionLayer matrix multiply routines
      *
-     * @param[in] input   Input tensor. Data types supported: QS8/QASYMM8/QS16/F16/F32.
-     * @param[in] weights Weights tensor. Data type supported: Same as @p input.
-     * @param[in] output  Output tensor. Data types supported: Same as @p input,
-     *                                      except for input of QASYMM8 type where output should be of S32 type.
+     * @param[in] input         Input tensor. Data types supported: QS8/QASYMM8/QS16/F16/F32.
+     * @param[in] weights       Weights tensor. Data type supported: Same as @p input.
+     * @param[in] output        Output tensor. Data types supported: Same as @p input,
+     *                          except for input of QASYMM8 type where output should be of S32 type.
+     * @param[in] gemm_3d_depth (Optional) Depth of GEMM 3D (Defaults to 1)
      *
      * @return a status
      */
-    static Status validate_mm(const ITensorInfo *input, const ITensorInfo *weights, const ITensorInfo *output);
+    static Status validate_mm(const ITensorInfo *input, const ITensorInfo *weights, const ITensorInfo *output, int gemm_3d_depth = 1);
 
 private:
     CLMemoryGroup                                       _memory_group;
@@ -192,9 +194,12 @@ private:
     CLTensor _gemm_output;
     CLTensor _tmp_output;
 
+    DataLayout _data_layout;
+
+    bool _skip_im2col;
     bool _is_quantized;
     bool _is_activationlayer_enabled;
     bool _is_prepared;
 };
-}
+} // namespace arm_compute
 #endif /* __ARM_COMPUTE_CLGEMMCONVOLUTIONLAYER_H__ */