diff options
| author | Pablo Marquez Tello <pablo.tello@arm.com> | 2022-03-17 12:52:02 +0000 |
|---|---|---|
| committer | ramelg01 <ramy.elgammal@arm.com> | 2022-03-29 10:26:17 +0100 |
| commit | a5d61bf5cd566955f3902e07c43c5c1c059bf8e9 (patch) | |
| tree | 7a5e4853d6998c207ca4f333df707d64068bb790 /src/cpu/kernels/CpuQuantizeKernel.h | |
| parent | b83e67238bf84c5780f9d27c87cf30342099b291 (diff) | |
| download | ComputeLibrary-a5d61bf5cd566955f3902e07c43c5c1c059bf8e9.tar.gz | |
NEQLSTM: Add support for QASYMM8_SIGNED for input_to_forget_weights
* QLSTM only supports QSYMM8 for the argument input_to_forget_weights
* We add support for QASYMM8_SIGNED by dequantizing and requantizing to QSYMM8
* Resolves COMPMID-5184
Change-Id: I1cae18d81dafdb7ae722b520a1354cf4a56b9606
Signed-off-by: Pablo Marquez Tello <pablo.tello@arm.com>
Reviewed-on: https://review.mlplatform.org/c/ml/ComputeLibrary/+/7321
Tested-by: Arm Jenkins <bsgcomp@arm.com>
Reviewed-by: Gunes Bayir <gunes.bayir@arm.com>
Comments-Addressed: Arm Jenkins <bsgcomp@arm.com>
(cherry picked from commit 187a041dedf8e9db0c9e0652f13f8639dca880f3)
Diffstat (limited to 'src/cpu/kernels/CpuQuantizeKernel.h')
| -rw-r--r-- | src/cpu/kernels/CpuQuantizeKernel.h | 3 |
1 files changed, 3 insertions, 0 deletions
diff --git a/src/cpu/kernels/CpuQuantizeKernel.h b/src/cpu/kernels/CpuQuantizeKernel.h index 28690bea54..2bc8105a11 100644 --- a/src/cpu/kernels/CpuQuantizeKernel.h +++ b/src/cpu/kernels/CpuQuantizeKernel.h @@ -81,6 +81,9 @@ private: template <typename T> void run_quantize_qasymm16(const ITensor *src, ITensor *dst, const Window &window); + template <typename TIn, typename TOut> + void run_quantize_qsymm8(const ITensor *src, ITensor *dst, const Window &window); + QuantizeFunctionExecutorPtr _func{ nullptr }; }; } // namespace kernels |