commit 70d43a3671090d7ab104909a9433c88e02593038
author Michalis Spyrou <michalis.spyrou@arm.com> Mon Jun 22 17:05:43 2020 +0100
committer Michalis Spyrou <michalis.spyrou@arm.com> Thu Jun 25 13:21:00 2020 +0000
tree 44394282795be5b17bebb65f228303dbf7600bfe
parent c41a6a611973cb245220641e06f8fa984b156954

COMPMID-3538: Remove templates from NEGEMMLowpOffsetContributionOutputStageKernel

This change reduces the core's library size by 191Kb.

Change-Id: Ifb8eb0d7f8bc7713f2368803a62a4c9277cc5c87
Signed-off-by: Michalis Spyrou <michalis.spyrou@arm.com>
Reviewed-on: https://review.mlplatform.org/c/ml/ComputeLibrary/+/3439
Reviewed-by: Michele Di Giorgio <michele.digiorgio@arm.com>
Reviewed-by: Georgios Pinitas <georgios.pinitas@arm.com>
Tested-by: Arm Jenkins <bsgcomp@arm.com>
Comments-Addressed: Arm Jenkins <bsgcomp@arm.com>

src/core/NEON/kernels/NEGEMMLowpQuantizeDownInt32ToUint8ScaleByFixedPointKernel.cpp

diff --git a/src/core/NEON/kernels/NEGEMMLowpQuantizeDownInt32ToUint8ScaleByFixedPointKernel.cpp b/src/core/NEON/kernels/NEGEMMLowpQuantizeDownInt32ToUint8ScaleByFixedPointKernel.cpp
index 4a9d2f7..78610c9 100644
--- a/src/core/NEON/kernels/NEGEMMLowpQuantizeDownInt32ToUint8ScaleByFixedPointKernel.cpp
+++ b/src/core/NEON/kernels/NEGEMMLowpQuantizeDownInt32ToUint8ScaleByFixedPointKernel.cpp
@@ -145,7 +145,7 @@
                 in_s32.val[2] = vaddq_s32(in_s32.val[2], bias_s32.val[2]);
                 in_s32.val[3] = vaddq_s32(in_s32.val[3], bias_s32.val[3]);
 
-                vst1q_u8(out.ptr() + x, finalize_quantization<is_bounded_relu>(in_s32, _result_fixedpoint_multiplier, _result_shift, result_offset_after_shift_s32, min_u8, max_u8));
+                vst1q_u8(out.ptr() + x, finalize_quantization(in_s32, _result_fixedpoint_multiplier, _result_shift, result_offset_after_shift_s32, min_u8, max_u8, is_bounded_relu));
             }
 
             // Compute left-over elements
@@ -157,7 +157,7 @@
                 // Add bias
                 in_value += bias_value;
                 // Finalize and store the result
-                *(out.ptr() + x) = finalize_quantization<is_bounded_relu>(in_value, _result_fixedpoint_multiplier, _result_shift, _result_offset_after_shift, static_cast<uint8_t>(_min), static_cast<uint8_t>(_max));
+                *(out.ptr() + x) = finalize_quantization(in_value, _result_fixedpoint_multiplier, _result_shift, _result_offset_after_shift, static_cast<uint8_t>(_min), static_cast<uint8_t>(_max), is_bounded_relu);
             }
         },
         in, out, bias);
@@ -180,7 +180,7 @@
                     }
                 };
 
-                vst1q_u8(out.ptr() + x, finalize_quantization<is_bounded_relu>(in_s32, _result_fixedpoint_multiplier, _result_shift, result_offset_after_shift_s32, min_u8, max_u8));
+                vst1q_u8(out.ptr() + x, finalize_quantization(in_s32, _result_fixedpoint_multiplier, _result_shift, result_offset_after_shift_s32, min_u8, max_u8, is_bounded_relu));
             }
 
             // Compute left-over elements
@@ -189,7 +189,7 @@
                 const int32_t in_value = *(reinterpret_cast<const int32_t *>(in.ptr()) + x);
 
                 // Finalize and store the result
-                *(out.ptr() + x) = finalize_quantization<is_bounded_relu>(in_value, _result_fixedpoint_multiplier, _result_shift, _result_offset_after_shift, static_cast<uint8_t>(_min), static_cast<uint8_t>(_max));
+                *(out.ptr() + x) = finalize_quantization(in_value, _result_fixedpoint_multiplier, _result_shift, _result_offset_after_shift, static_cast<uint8_t>(_min), static_cast<uint8_t>(_max), is_bounded_relu);
             }
         },
         in, out);
@@ -244,4 +244,4 @@
 
     (this->*_func)(window);
 }
-} // namespace arm_compute
\ No newline at end of file
+} // namespace arm_compute