commit 959c26d0457deeebf7306b9e4317863f144415b5
author Manuel Bottini <manuel.bottini@arm.com> Mon Dec 02 16:22:35 2019 +0000
committer Manuel Bottini <manuel.bottini@arm.com> Tue Jan 14 13:15:11 2020 +0000
tree 9a439d27b9985f21b3b1b27db519efe9e928954a
parent 6427c8233661f81053d1ad486b5914c612cef3d6

COMPMID-2790: Add support for QASYMM8_SIGNED in CLGEMMLowpMatrixMultiplyCore

Change-Id: Ifdaeb53c512ba697f174649c026075010f54f628
Signed-off-by: Manuel Bottini <manuel.bottini@arm.com>
Reviewed-on: https://review.mlplatform.org/c/2472
Reviewed-by: Michele Di Giorgio <michele.digiorgio@arm.com>
Comments-Addressed: Arm Jenkins <bsgcomp@arm.com>
Reviewed-by: Sang-Hoon Park <sang-hoon.park@arm.com>
Tested-by: Arm Jenkins <bsgcomp@arm.com>
Reviewed-by: Giuseppe Rossini <giuseppe.rossini@arm.com>

src/runtime/CL/functions/CLGEMMLowpMatrixMultiplyCore.cpp

diff --git a/src/runtime/CL/functions/CLGEMMLowpMatrixMultiplyCore.cpp b/src/runtime/CL/functions/CLGEMMLowpMatrixMultiplyCore.cpp
index 4c0a521..cdb78c2 100644
--- a/src/runtime/CL/functions/CLGEMMLowpMatrixMultiplyCore.cpp
+++ b/src/runtime/CL/functions/CLGEMMLowpMatrixMultiplyCore.cpp
@@ -1,5 +1,5 @@
 /*
- * Copyright (c) 2017-2019 ARM Limited.
+ * Copyright (c) 2017-2020 ARM Limited.
  *
  * SPDX-License-Identifier: MIT
  *
@@ -206,8 +206,10 @@
         _gemm_output_stage_multipliers.allocator()->init(TensorInfo(TensorShape(num_filters), 1, DataType::S32));
         _gemm_output_stage_shifts.allocator()->init(TensorInfo(TensorShape(num_filters), 1, DataType::S32));
 
+        GEMMLowpOutputStageInfo gemmlowp_output_stage = gemm_info.gemmlowp_output_stage();
+        gemmlowp_output_stage.output_data_type        = _matrix_a->info()->data_type();
         _offset_contribution_output_stage_kernel.configure(&_mm_result_s32, _a_offset == 0 ? nullptr : &_vector_sum_col, _b_offset == 0 ? nullptr : &_vector_sum_row, c, output, a->info()->dimension(0),
-                                                           _a_offset, _b_offset, gemm_info.gemmlowp_output_stage(), &_gemm_output_stage_multipliers, &_gemm_output_stage_shifts);
+                                                           _a_offset, _b_offset, gemmlowp_output_stage, &_gemm_output_stage_multipliers, &_gemm_output_stage_shifts);
 
         _gemm_output_stage_multipliers.allocator()->allocate();
         _gemm_output_stage_shifts.allocator()->allocate();
@@ -271,13 +273,10 @@
 
 Status CLGEMMLowpMatrixMultiplyCore::validate(const ITensorInfo *a, const ITensorInfo *b, const ITensorInfo *c, const ITensorInfo *output, const GEMMInfo &gemm_info)
 {
-    ARM_COMPUTE_RETURN_ERROR_ON_DATA_TYPE_CHANNEL_NOT_IN(a, 1, DataType::QASYMM8);
-    if(b->data_type() == DataType::QSYMM8_PER_CHANNEL)
-    {
-        //DataType::QSYMM8_PER_CHANNEL supported only for weights
-        ARM_COMPUTE_RETURN_ERROR_ON_MSG(a->data_type() != DataType::QASYMM8, "Matrix A is not quantized while Matrix B is");
-    }
-    else
+    ARM_COMPUTE_ERROR_ON_NULLPTR(a, b, output);
+    ARM_COMPUTE_RETURN_ERROR_ON_DATA_TYPE_CHANNEL_NOT_IN(a, 1, DataType::QASYMM8, DataType::QASYMM8_SIGNED);
+    //DataType::QSYMM8_PER_CHANNEL supported only for weights
+    if(b->data_type() != DataType::QSYMM8_PER_CHANNEL)
     {
         ARM_COMPUTE_RETURN_ERROR_ON_MISMATCHING_DATA_TYPES(a, b);
     }
@@ -388,13 +387,15 @@
 
         const TensorInfo gemm_output_stage_multipliers_shifts_info(TensorInfo(TensorShape(num_filters), 1, DataType::S32));
 
+        GEMMLowpOutputStageInfo gemmlowp_output_stage = gemm_info.gemmlowp_output_stage();
+        gemmlowp_output_stage.output_data_type        = a->data_type();
         ARM_COMPUTE_RETURN_ON_ERROR(CLGEMMLowpOffsetContributionOutputStageKernel::validate(&mm_result_s32_info,
                                                                                             a_offset == 0 ? nullptr : &info_vector_sum_col,
                                                                                             b_offset == 0 ? nullptr : &info_vector_sum_row,
                                                                                             c,
                                                                                             output,
                                                                                             a_offset, b_offset,
-                                                                                            gemm_info.gemmlowp_output_stage(),
+                                                                                            gemmlowp_output_stage,
                                                                                             &gemm_output_stage_multipliers_shifts_info,
                                                                                             &gemm_output_stage_multipliers_shifts_info));
     }