Implement OpenCL MatMul for Lhs NT Rhs T/NT FP32/16

 - Implement ClNativeMatMulKernel class
 - Implement opencl kernel for LHS non-transposed and RHS non-transposed
 - Implement opencl kernel for LHS non-transposed and RHS transposed
 - Add test fixture and dataset for matmul
 - Implement transpose_tensor() for reference implementation to transpose high dimensional tensors

Resolves: COMPMID-5944, COMPMID-5951

Co-authored-by: Gunes Bayir <gunes.bayir@arm.com>
Co-authored-by: Ramy Elgammal <ramy.elgammal@arm.com>
Change-Id: I1d5b8978f41be27baddb3153ade880472141573f
Signed-off-by: Gunes Bayir <gunes.bayir@arm.com>
Signed-off-by: Ramy Elgammal <ramy.elgammal@arm.com>
Reviewed-on: https://review.mlplatform.org/c/ml/ComputeLibrary/+/9333
Tested-by: Arm Jenkins <bsgcomp@arm.com>
Reviewed-by: Gian Marco Iodice <gianmarco.iodice@arm.com>
Benchmark: Arm Jenkins <bsgcomp@arm.com>

2 files changed, 48 insertions, 6 deletions

diff --git a/arm_compute/core/KernelDescriptors.h b/arm_compute/core/KernelDescriptors.h
index 19ac254c04..016e03d88e 100644
--- a/arm_compute/core/KernelDescriptors.h
+++ b/arm_compute/core/KernelDescriptors.h
@@ -21,8 +21,8 @@
  * OUT OF OR IN CONNECTION WITH THE SOFTWARE OR THE USE OR OTHER DEALINGS IN THE
  * SOFTWARE.
  */
-#ifndef ARM_COMPUTE_CORE_KERNEL_DESCRIPTORS_H
-#define ARM_COMPUTE_CORE_KERNEL_DESCRIPTORS_H
+#ifndef ACL_ARM_COMPUTE_CORE_KERNELDESCRIPTORS
+#define ACL_ARM_COMPUTE_CORE_KERNELDESCRIPTORS
 
 #include "arm_compute/core/PixelValue.h"
 #include "arm_compute/core/Types.h"
@@ -223,5 +223,19 @@ struct ScaleKernelInfo
     bool                align_corners;         /**< Align corners of input and output */
     DataLayout          data_layout;           /**< Data layout to use */
 };
+
+struct MatMulKernelInfo
+{
+    MatMulKernelInfo(bool adj_lhs = false, bool adj_rhs = false, int m0 = 1, int n0 = 1, int k0 = 1, bool export_rhs_to_cl_image = false)
+        : adj_lhs{ adj_lhs }, adj_rhs{ adj_rhs }, m0{ m0 }, n0{ n0 }, k0{ k0 }, export_rhs_to_cl_image{ export_rhs_to_cl_image }
+    {
+    }
+    bool adj_lhs{ false };                /**< Get Adjoint LHS flag value */
+    bool adj_rhs{ false };                /**< Get Adjoint RHS flag value */
+    int  m0{ 1 };                         /**< Number of output rows processed by each work-item*/
+    int  n0{ 1 };                         /**< Number of output columns processed by each work-item*/
+    int  k0{ 1 };                         /**< Number of inner accumulations */
+    bool export_rhs_to_cl_image{ false }; /**< Flag to know whether the RHS tensor should be exported to cl_image*/
+};
 } // namespace arm_compute
-#endif /* ARM_COMPUTE_CORE_KERNEL_DESCRIPTORS_H */
+#endif /* ACL_ARM_COMPUTE_CORE_KERNELDESCRIPTORS */
diff --git a/arm_compute/core/utils/misc/ShapeCalculator.h b/arm_compute/core/utils/misc/ShapeCalculator.h
index 6655cc1439..75a063f75c 100644
--- a/arm_compute/core/utils/misc/ShapeCalculator.h
+++ b/arm_compute/core/utils/misc/ShapeCalculator.h
@@ -21,8 +21,8 @@
  * OUT OF OR IN CONNECTION WITH THE SOFTWARE OR THE USE OR OTHER DEALINGS IN THE
  * SOFTWARE.
  */
-#ifndef ARM_COMPUTE_MISC_SHAPE_CALCULATOR_H
-#define ARM_COMPUTE_MISC_SHAPE_CALCULATOR_H
+#ifndef ACL_ARM_COMPUTE_CORE_UTILS_MISC_SHAPECALCULATOR
+#define ACL_ARM_COMPUTE_CORE_UTILS_MISC_SHAPECALCULATOR
 
 #include "arm_compute/core/Helpers.h"
 #include "arm_compute/core/ITensorInfo.h"
@@ -1008,6 +1008,34 @@ inline TensorShape compute_mm_shape(const ITensorInfo &input0, const ITensorInfo
 
 /** Calculate the matrix multiplication output shape of two tensors
  *
+ * @param[in] input0      First input tensor info
+ * @param[in] input1      Second input tensor info
+ * @param[in] matmul_info Batch MatMul Kernel info to know which matrix is transposed
+ *
+ * @return the calculated shape
+ */
+inline TensorShape compute_batchmatmul_shape(const TensorShape &input0, const TensorShape &input1, const MatMulKernelInfo &matmul_info)
+{
+    TensorShape output_shape{ input0 };
+
+    if(matmul_info.adj_lhs)
+    {
+        output_shape.set(1, input0[0]); // The vertical (M) dimension
+    }
+
+    if(matmul_info.adj_rhs)
+    {
+        output_shape.set(0, input1[1]); // The horizontal (N) dimension
+    }
+    else
+    {
+        output_shape.set(0, input1[0]); // The horizontal (N) dimension
+    }
+
+    return output_shape;
+}
+/** Calculate the matrix multiplication output shape of two tensors
+ *
  * @param[in] input           Input tensor info
  * @param[in] gemm_3d_depth   (Optional)  GEMM 3d depth
  * @param[in] batch_size_on_z (Optional) True if batch size is on z axis
@@ -1579,4 +1607,4 @@ inline TensorShape compute_gather_shape(const TensorShape &input_shape, const Te
 } // namespace shape_calculator
 } // namespace misc
 } // namespace arm_compute
-#endif /* ARM_COMPUTE_MISC_SHAPE_CALCULATOR_H */
+#endif /* ACL_ARM_COMPUTE_CORE_UTILS_MISC_SHAPECALCULATOR */