commit 448a81fcec04333364a1e3266d5081596d3a0477
author Georgios Pinitas <georgios.pinitas@arm.com> Thu Nov 21 14:10:25 2019 +0000
committer Georgios Pinitas <georgios.pinitas@arm.com> Wed Nov 27 10:56:10 2019 +0000
tree bd5382a58fae39a8014157423a8ff339d39e14b9
parent 449cbf9c20287fca9a56898cdc5821c061a66ce3

COMPMID-2805: Add QASYMM8_SIGNED support in NEGEMMLowpOutputStage

Add support from requantizing down from S32 to Int8 with fixed point
requantization. This involves the following:
- Compute fixed point multiplication between each entry of input by
  result_fixedpoint_multiplier
- Add bias to final result if bias tensor is not a nullptr
- Round to nearest division by a power-of-two using result_shift
- Add offset to each result
- Clamp the value between the specified min and max bounds
- Cast to int8 data type

Change-Id: I641b3fac0833c568d8565ccb859bbc561a24c17d
Signed-off-by: Georgios Pinitas <georgios.pinitas@arm.com>
Reviewed-on: https://review.mlplatform.org/c/2340
Comments-Addressed: Arm Jenkins <bsgcomp@arm.com>
Reviewed-by: Michele Di Giorgio <michele.digiorgio@arm.com>
Tested-by: Arm Jenkins <bsgcomp@arm.com>

tests/validation/reference/GEMMLowp.h

diff --git a/tests/validation/reference/GEMMLowp.h b/tests/validation/reference/GEMMLowp.h
index 815527e..7ff01ef 100644
--- a/tests/validation/reference/GEMMLowp.h
+++ b/tests/validation/reference/GEMMLowp.h
@@ -52,20 +52,13 @@
 SimpleTensor<uint8_t> gemmlowp_quantize_down_int32_to_uint8_scale(const SimpleTensor<T> &in, const SimpleTensor<T> &bias, int32_t result_offset, std::vector<int32_t> result_mult_int,
                                                                   std::vector<int32_t> result_shift, int32_t min = 0, int32_t max = 0);
 
-template <typename T>
-SimpleTensor<uint8_t> gemmlowp_quantize_down_int32_to_uint8_scale_by_fixedpoint(const SimpleTensor<T> &in, std::vector<int32_t> result_fixedpoint_multiplier, std::vector<int32_t> result_shift,
-                                                                                int32_t result_offset_after_shift, int32_t min = 0, int32_t max = 0);
+template <typename TIn, typename TOut>
+SimpleTensor<TOut> gemmlowp_quantize_down_scale_by_fixedpoint(const SimpleTensor<TIn> &in, std::vector<int32_t> result_fixedpoint_multiplier, std::vector<int32_t> result_shift,
+                                                              int32_t result_offset_after_shift, int32_t min = 0, int32_t max = 0);
 
-template <typename T>
-SimpleTensor<uint8_t> gemmlowp_quantize_down_int32_to_uint8_scale_by_fixedpoint(const SimpleTensor<T> &in, const SimpleTensor<T> &bias, std::vector<int32_t> result_fixedpoint_multiplier,
-                                                                                std::vector<int32_t> result_shift, int32_t result_offset_after_shift, int32_t min = 0, int32_t max = 0);
-
-template <typename T>
-SimpleTensor<int16_t> gemmlowp_quantize_down_int32_to_int16_scale_by_fixedpoint(const SimpleTensor<T> &in, int32_t result_fixedpoint_multiplier, int32_t result_shift,
-                                                                                int32_t min, int32_t max);
-template <typename T>
-SimpleTensor<int16_t> gemmlowp_quantize_down_int32_to_int16_scale_by_fixedpoint(const SimpleTensor<T> &in, const SimpleTensor<T> &bias, int32_t result_fixedpoint_multiplier,
-                                                                                int32_t result_shift, int32_t min, int32_t max);
+template <typename TIn, typename TOut>
+SimpleTensor<TOut> gemmlowp_quantize_down_scale_by_fixedpoint(const SimpleTensor<TIn> &in, const SimpleTensor<TIn> &bias, std::vector<int32_t> result_fixedpoint_multiplier,
+                                                              std::vector<int32_t> result_shift, int32_t result_offset_after_shift, int32_t min = 0, int32_t max = 0);
 } // namespace reference
 } // namespace validation
 } // namespace test