diff options
| author | Manuel Bottini <manuel.bottini@arm.com> | 2021-06-17 17:18:45 +0100 |
|---|---|---|
| committer | Manuel Bottini <manuel.bottini@arm.com> | 2021-06-22 17:03:54 +0000 |
| commit | ae58bdf3b58739e105a24e3640d0245e81cea5ee (patch) | |
| tree | e993b8768c3eff364a7c706db411c799fa86bfe0 /src/runtime/NEON/functions/NELSTMLayerQuantized.cpp | |
| parent | 2db3a9955ef22be4be8ccd5a45bc0973ef80e42a (diff) | |
| download | ComputeLibrary-ae58bdf3b58739e105a24e3640d0245e81cea5ee.tar.gz | |
Port NEGEMMLowp Part 1
Details:
Port NEGEMMLowpQuantizeDownInt32ScaleKernel to CpuGemmLowpQuantizeDownInt32ScaleKernel
Port NEGEMMLowpQuantizeDownInt32ToInt16ScaleByFixedPointKernel to CpuGemmLowpQuantizeDownInt32ToInt16ScaleByFixedPointKernel
Port NEGEMMLowpQuantizeDownInt32ToInt8ScaleByFixedPointKernel to CpuGemmLowpQuantizeDownInt32ToInt8ScaleByFixedPointKernel
Port NEGEMMLowpQuantizeDownInt32ToUint8ScaleByFixedPointKernel to CpuGemmLowpQuantizeDownInt32ToUint8ScaleByFixedPointKernel
Port NEGEMMLowpOutputStage functions to CpuGemmLowpOutputStage operators
Partially Resolves: COMPMID-4403
Change-Id: I6d5f45e43f35d731d564ed3b5c0e804d2a318fb1
Signed-off-by: Manuel Bottini <manuel.bottini@arm.com>
Reviewed-on: https://review.mlplatform.org/c/ml/ComputeLibrary/+/5833
Tested-by: Arm Jenkins <bsgcomp@arm.com>
Reviewed-by: Michele Di Giorgio <michele.digiorgio@arm.com>
Comments-Addressed: Arm Jenkins <bsgcomp@arm.com>
Diffstat (limited to 'src/runtime/NEON/functions/NELSTMLayerQuantized.cpp')
| -rw-r--r-- | src/runtime/NEON/functions/NELSTMLayerQuantized.cpp | 15 |
1 files changed, 13 insertions, 2 deletions
diff --git a/src/runtime/NEON/functions/NELSTMLayerQuantized.cpp b/src/runtime/NEON/functions/NELSTMLayerQuantized.cpp index 5c0f19a15c..a56e16864a 100644 --- a/src/runtime/NEON/functions/NELSTMLayerQuantized.cpp +++ b/src/runtime/NEON/functions/NELSTMLayerQuantized.cpp @@ -152,7 +152,13 @@ void NELSTMLayerQuantized::configure(const ITensor *input, quantization::calculate_quantized_multiplier(multiplier, &output_multiplier, &output_shift); _memory_group.manage(&_output_lowp); - _output_stage.configure(&_output_highp, &_bias, &_output_lowp, output_multiplier, output_shift); + + GEMMLowpOutputStageInfo info; + info.type = GEMMLowpOutputStageType::QUANTIZE_DOWN_FIXEDPOINT; + info.gemmlowp_multiplier = output_multiplier; + info.gemmlowp_shift = output_shift; + info.output_data_type = DataType::QSYMM16; + _output_stage.configure(&_output_highp, &_bias, &_output_lowp, info); _output_highp.allocator()->allocate(); _bias.allocator()->allocate(); @@ -358,7 +364,12 @@ Status NELSTMLayerQuantized::validate(const ITensorInfo *input, ARM_COMPUTE_RETURN_ON_ERROR(quantization::calculate_quantized_multiplier(multiplier, &output_multiplier, &output_shift)); // _output_stage - ARM_COMPUTE_RETURN_ON_ERROR(NEGEMMLowpQuantizeDownInt32ToInt16ScaleByFixedPoint::validate(&output_highp, &bias_concatenated, &output_lowp)); + GEMMLowpOutputStageInfo info; + info.type = GEMMLowpOutputStageType::QUANTIZE_DOWN_FIXEDPOINT; + info.gemmlowp_multiplier = output_multiplier; + info.gemmlowp_shift = output_shift; + info.output_data_type = DataType::QSYMM16; + ARM_COMPUTE_RETURN_ON_ERROR(NEGEMMLowpOutputStage::validate(&output_highp, &bias_concatenated, &output_lowp, info)); TensorInfo input_gate_input; TensorInfo forget_gate_input; |