COMPMID-1698: Implementing CLGEMMLowpMatrixMultiplyReshapedKernel

Change-Id: Ia4db21b394a0b9235393202ce3c00b11cceb94ea
Reviewed-on: https://review.mlplatform.org/568
Tested-by: Arm Jenkins <bsgcomp@arm.com>
Reviewed-by: Michele Di Giorgio <michele.digiorgio@arm.com>

5 files changed, 96 insertions, 4 deletions

diff --git a/arm_compute/core/CL/CLKernels.h b/arm_compute/core/CL/CLKernels.h
index e68769b6ae..07e214be3f 100644
--- a/arm_compute/core/CL/CLKernels.h
+++ b/arm_compute/core/CL/CLKernels.h
@@ -69,6 +69,7 @@
 #include "arm_compute/core/CL/kernels/CLFuseBatchNormalizationKernel.h"
 #include "arm_compute/core/CL/kernels/CLGEMMInterleave4x4Kernel.h"
 #include "arm_compute/core/CL/kernels/CLGEMMLowpMatrixMultiplyKernel.h"
+#include "arm_compute/core/CL/kernels/CLGEMMLowpMatrixMultiplyReshapedKernel.h"
 #include "arm_compute/core/CL/kernels/CLGEMMLowpOffsetContributionKernel.h"
 #include "arm_compute/core/CL/kernels/CLGEMMLowpOffsetContributionOutputStageKernel.h"
 #include "arm_compute/core/CL/kernels/CLGEMMLowpQuantizeDownInt32ToUint8ScaleByFixedPointKernel.h"
diff --git a/arm_compute/core/CL/kernels/CLGEMMLowpMatrixMultiplyReshapedKernel.h b/arm_compute/core/CL/kernels/CLGEMMLowpMatrixMultiplyReshapedKernel.h
new file mode 100644
index 0000000000..1cf7236446
--- /dev/null
+++ b/arm_compute/core/CL/kernels/CLGEMMLowpMatrixMultiplyReshapedKernel.h
@@ -0,0 +1,86 @@
+/*
+ * Copyright (c) 2019 ARM Limited.
+ *
+ * SPDX-License-Identifier: MIT
+ *
+ * Permission is hereby granted, free of charge, to any person obtaining a copy
+ * of this software and associated documentation files (the "Software"), to
+ * deal in the Software without restriction, including without limitation the
+ * rights to use, copy, modify, merge, publish, distribute, sublicense, and/or
+ * sell copies of the Software, and to permit persons to whom the Software is
+ * furnished to do so, subject to the following conditions:
+ *
+ * The above copyright notice and this permission notice shall be included in all
+ * copies or substantial portions of the Software.
+ *
+ * THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS OR
+ * IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF MERCHANTABILITY,
+ * FITNESS FOR A PARTICULAR PURPOSE AND NONINFRINGEMENT. IN NO EVENT SHALL THE
+ * AUTHORS OR COPYRIGHT HOLDERS BE LIABLE FOR ANY CLAIM, DAMAGES OR OTHER
+ * LIABILITY, WHETHER IN AN ACTION OF CONTRACT, TORT OR OTHERWISE, ARISING FROM,
+ * OUT OF OR IN CONNECTION WITH THE SOFTWARE OR THE USE OR OTHER DEALINGS IN THE
+ * SOFTWARE.
+ */
+#ifndef __ARM_COMPUTE_CLGEMMLOWPMATRIXMULTIPLYRESHAPEDKERNEL_H__
+#define __ARM_COMPUTE_CLGEMMLOWPMATRIXMULTIPLYRESHAPEDKERNEL_H__
+
+#include "arm_compute/core/CL/ICLKernel.h"
+
+namespace arm_compute
+{
+class ICLTensor;
+
+/** OpenCL kernel to multiply matrices when both the input matrices LHS (input0) and RHS (input1) have been reshaped
+ *
+ * @note The input matrices @p input0 and @p input1 must be reshaped through @ref CLGEMMReshapeLHSMatrixKernel and  @ref CLGEMMReshapeRHSMatrixKernel
+ */
+class CLGEMMLowpMatrixMultiplyReshapedKernel : public ICLKernel
+{
+public:
+    /** Default Constructor */
+    CLGEMMLowpMatrixMultiplyReshapedKernel();
+    /** Prevent instances of this class from being copied (As this class contains pointers) */
+    CLGEMMLowpMatrixMultiplyReshapedKernel(const CLGEMMLowpMatrixMultiplyReshapedKernel &) = delete;
+    /** Prevent instances of this class from being copied (As this class contains pointers) */
+    CLGEMMLowpMatrixMultiplyReshapedKernel &operator=(const CLGEMMLowpMatrixMultiplyReshapedKernel &) = delete;
+    /** Allow instances of this class to be moved */
+    CLGEMMLowpMatrixMultiplyReshapedKernel(CLGEMMLowpMatrixMultiplyReshapedKernel &&) = default;
+    /** Allow instances of this class to be moved */
+    CLGEMMLowpMatrixMultiplyReshapedKernel &operator=(CLGEMMLowpMatrixMultiplyReshapedKernel &&) = default;
+    /** Initialise the kernel's input and output.
+     *
+     * @param[in]  input0    Input tensor containing the LHS reshaped matrix. Data type supported: QASYMM8
+     * @param[in]  input1    Input tensor containing the RHS reshaped matrix. Data type supported: same as @p input0
+     * @param[out] output    Output tensor to store the result of matrix multiplication. Data type supported: same as @p input0
+     * @param[in]  lhs_info  LHS matrix information used for reshaping the input0 tensor
+     * @param[in]  rhs_info  RHS matrix information used for reshaping the input1 tensor
+     * @param[in]  gemm_info GEMM information used to retrieve the original dimensions of the input matrices
+     */
+    void configure(const ICLTensor *input0, const ICLTensor *input1, ICLTensor *output, const GEMMLHSMatrixInfo &lhs_info, const GEMMRHSMatrixInfo &rhs_info, const GEMMReshapeInfo &gemm_info);
+    /** Static function to check if given info will lead to a valid configuration of @ref CLGEMMLowpMatrixMultiplyReshapedKernel
+     *
+     * @param[in] input0    Input tensor info containing the LHS reshaped matrix. Data type supported: QASYMM8
+     * @param[in] input1    Input tensor info containing the RHS reshaped matrix. Data type supported: same as @p input0
+     * @param[in] output    Output tensor info. Data type supported: same as @p input0
+     * @param[in] lhs_info  LHS matrix information used for reshaping the input0 tensor
+     * @param[in] rhs_info  RHS matrix information used for reshaping the input1 tensor
+     * @param[in] gemm_info GEMM information used to retrieve the original dimensions of the input matrices
+     *
+     * @return a status
+     */
+    static Status validate(const ITensorInfo *input0, const ITensorInfo *input1, const ITensorInfo *output, const GEMMLHSMatrixInfo &lhs_info, const GEMMRHSMatrixInfo &rhs_info,
+                           const GEMMReshapeInfo &gemm_info);
+
+    // Inherited methods overridden:
+    void run(const Window &window, cl::CommandQueue &queue) override;
+
+private:
+    const ICLTensor *_input0;
+    const ICLTensor *_input1;
+    ICLTensor       *_output;
+    bool             _slide_matrix_b;
+    bool             _reinterpret_output_as_3d;
+    unsigned int     _k;
+};
+} // namespace arm_compute
+#endif /*__ARM_COMPUTE_CLGEMMLOWPMATRIXMULTIPLYRESHAPEDKERNEL_H__*/
\ No newline at end of file
diff --git a/arm_compute/runtime/CL/functions/CLGEMM.h b/arm_compute/runtime/CL/functions/CLGEMM.h
index 3ec07cf5f9..0bad446551 100644
--- a/arm_compute/runtime/CL/functions/CLGEMM.h
+++ b/arm_compute/runtime/CL/functions/CLGEMM.h
@@ -115,7 +115,7 @@ private:
     bool                               _run_addition;
     bool                               _reshape_b_only_on_first_run;
     bool                               _is_prepared;
-    bool                               _is_new_gemm_reshaped; // Removed when COMPMID-1892 is completed
+    bool                               _is_new_gemm_reshaped; // Remove when COMPMID-1892 is completed
 };
 } // namespace arm_compute
 
diff --git a/arm_compute/runtime/CL/functions/CLGEMMLowpMatrixMultiplyCore.h b/arm_compute/runtime/CL/functions/CLGEMMLowpMatrixMultiplyCore.h
index 72d91070f8..4345ff267b 100644
--- a/arm_compute/runtime/CL/functions/CLGEMMLowpMatrixMultiplyCore.h
+++ b/arm_compute/runtime/CL/functions/CLGEMMLowpMatrixMultiplyCore.h
@@ -25,6 +25,7 @@
 #define __ARM_COMPUTE_CLGEMMLOWPMATRIXMULTIPLYCORE_H__
 
 #include "arm_compute/core/CL/kernels/CLGEMMLowpMatrixMultiplyKernel.h"
+#include "arm_compute/core/CL/kernels/CLGEMMLowpMatrixMultiplyReshapedKernel.h"
 #include "arm_compute/core/CL/kernels/CLGEMMLowpOffsetContributionKernel.h"
 #include "arm_compute/core/CL/kernels/CLGEMMLowpOffsetContributionOutputStageKernel.h"
 #include "arm_compute/core/CL/kernels/CLGEMMLowpReductionKernel.h"
@@ -43,7 +44,8 @@ class ICLTensor;
  *
  *  -# @ref CLGEMMReshapeLHSMatrixKernel  (if the output tensor is a matrix)
  *  -# @ref CLGEMMReshapeRHSMatrixKernel  (if the output tensor is a matrix)
- *  -# @ref CLGEMMLowpMatrixMultiplyKernel
+ *  -# @ref CLGEMMLowpMatrixMultiplyKernel (if the input matrix is a vector or for Midgard architectures)
+ *  -# @ref CLGEMMLowpMatrixMultiplyReshapedKernel (if the input matrix is not a vector and if the GPU architecture is not Midgard)
  *  -# @ref CLGEMMLowpMatrixAReductionKernel (if the offset of matrix B is not 0)
  *  -# @ref CLGEMMLowpMatrixBReductionKernel (if the offset of matrix A is not 0)
  *  -# @ref CLGEMMLowpOffsetContributionKernel (if gemm_info.gemmlowp_output_stage == NONE)
@@ -101,6 +103,7 @@ public:
 private:
     CLMemoryGroup                                 _memory_group;
     CLGEMMLowpMatrixMultiplyKernel                _mm_kernel;
+    CLGEMMLowpMatrixMultiplyReshapedKernel        _mm_reshaped_kernel;
     CLGEMMReshapeLHSMatrixKernel                  _mtx_a_reshape_kernel;
     CLGEMMReshapeRHSMatrixKernel                  _mtx_b_reshape_kernel;
     CLGEMMLowpMatrixAReductionKernel              _mtx_a_reduction_kernel;
@@ -115,10 +118,10 @@ private:
     const ICLTensor                              *_original_b;
     int32_t                                       _a_offset;
     int32_t                                       _b_offset;
-    bool                                          _is_interleaved_transposed;
+    bool                                          _is_gemm_reshaped;
     bool                                          _reshape_b_only_on_first_run;
     bool                                          _is_prepared;
     bool                                          _fuse_output_stage;
 };
 }
-#endif /*__ARM_COMPUTE_CLGEMMLOWPMATRIXMULTIPLYCORE_H__ */
+#endif /*__ARM_COMPUTE_CLGEMMLOWPMATRIXMULTIPLYCORE_H__ */
\ No newline at end of file
diff --git a/arm_compute/runtime/CL/gemm_reshaped/CLGEMMReshapedConfigurationBifrost.h b/arm_compute/runtime/CL/gemm_reshaped/CLGEMMReshapedConfigurationBifrost.h
index b452c53c39..c452e159cf 100644
--- a/arm_compute/runtime/CL/gemm_reshaped/CLGEMMReshapedConfigurationBifrost.h
+++ b/arm_compute/runtime/CL/gemm_reshaped/CLGEMMReshapedConfigurationBifrost.h
@@ -40,6 +40,8 @@ public:
 private:
     std::pair<GEMMLHSMatrixInfo, GEMMRHSMatrixInfo> configure_G7x_f32(unsigned int m, unsigned int n, unsigned int k, unsigned int b);
     std::pair<GEMMLHSMatrixInfo, GEMMRHSMatrixInfo> configure_G76_f32(unsigned int m, unsigned int n, unsigned int k, unsigned int b);
+    std::pair<GEMMLHSMatrixInfo, GEMMRHSMatrixInfo> configure_G7x_u8(unsigned int m, unsigned int n, unsigned int k, unsigned int b);
+    std::pair<GEMMLHSMatrixInfo, GEMMRHSMatrixInfo> configure_G76_u8(unsigned int m, unsigned int n, unsigned int k, unsigned int b);
 };
 } // namespace cl_gemm
 } // namespace arm_compute