Add per-channel quantization support for CLDeconvolutionLayer
Add QSYMM8_PER_CHANNEL support on weight input for CLDeconvolutionLayer.
When weights are per-channel quantized type "Direct" method is always
used.
Also reduce number of QSYMM8_PER_CHANNEL tests for NEDeconvolutionLayer.
Resolves: COMPMID-3438
Signed-off-by: Freddie Liardet <frederick.liardet@arm.com>
Change-Id: I1330cac5142e19d21e322574fb8d912558745b02
Reviewed-on: https://review.mlplatform.org/c/ml/ComputeLibrary/+/5484
Reviewed-by: Michele Di Giorgio <michele.digiorgio@arm.com>
Reviewed-by: Giorgio Arena <giorgio.arena@arm.com>
Tested-by: Arm Jenkins <bsgcomp@arm.com>
Comments-Addressed: Arm Jenkins <bsgcomp@arm.com>
diff --git a/arm_compute/runtime/CL/functions/CLDeconvolutionLayer.h b/arm_compute/runtime/CL/functions/CLDeconvolutionLayer.h
index df3cad6..4be8c17 100644
--- a/arm_compute/runtime/CL/functions/CLDeconvolutionLayer.h
+++ b/arm_compute/runtime/CL/functions/CLDeconvolutionLayer.h
@@ -1,5 +1,5 @@
/*
- * Copyright (c) 2017-2020 Arm Limited.
+ * Copyright (c) 2017-2021 Arm Limited.
*
* SPDX-License-Identifier: MIT
*
@@ -47,8 +47,8 @@
/** Set the input, weights, biases and output tensors.
*
* @param[in,out] input Input tensor. 3 lower dimensions represent a single input, and an optional 4th dimension for batch of inputs. Data types supported: QASYMM8_SIGNED/QASYMM8/F16/F32.
- * @param[in] weights The 4d weights with dimensions [width, height, IFM, OFM]. Data type supported: Same as @p input.
- * @param[in] bias (Optional) The biases have one dimension. Data type supported: Same as @p input.
+ * @param[in] weights The 4d weights with dimensions [width, height, IFM, OFM]. Data type supported: Same as @p input or QSYMM8_PER_CHANNEL if @p input is QASYMM8/QASYMM8_SIGNED.
+ * @param[in] bias (Optional) The biases have one dimension. Data type supported: Should match @p input data type, except for input of QASYMM8 and QASYMM8_SIGNED type where biases should be of S32 type
* @param[out] output Output tensor. The output has the same number of dimensions as the @p input.
* @param[in] deconv_info Contains padding and policies to be used in the deconvolution, this is described in @ref PadStrideInfo.
* @param[in] weights_info (Optional) Weights information needed for @ref CLConvolutionLayer, specifies if the weights tensor has been reshaped with @ref CLWeightsReshapeKernel.
@@ -59,8 +59,8 @@
*
* @param[in] compile_context The compile context to be used.
* @param[in,out] input Input tensor. 3 lower dimensions represent a single input, and an optional 4th dimension for batch of inputs. Data types supported: QASYMM8_SIGNED/QASYMM8/F16/F32.
- * @param[in] weights The 4d weights with dimensions [width, height, IFM, OFM]. Data type supported: Same as @p input.
- * @param[in] bias (Optional) The biases have one dimension. Data type supported: Same as @p input.
+ * @param[in] weights The 4d weights with dimensions [width, height, IFM, OFM]. Data type supported: Same as @p input or QSYMM8_PER_CHANNEL if @p input is QASYMM8/QASYMM8_SIGNED.
+ * @param[in] bias (Optional) The biases have one dimension. Data type supported: Should match @p input data type, except for input of QASYMM8 and QASYMM8_SIGNED type where biases should be of S32 type
* @param[out] output Output tensor. The output has the same number of dimensions as the @p input.
* @param[in] deconv_info Contains padding and policies to be used in the deconvolution, this is described in @ref PadStrideInfo.
* @param[in] weights_info (Optional) Weights information needed for @ref CLConvolutionLayer, specifies if the weights tensor has been reshaped with @ref CLWeightsReshapeKernel.
@@ -71,8 +71,8 @@
/** Static function to check if given info will lead to a valid configuration of @ref CLDeconvolutionLayer
*
* @param[in] input Input tensor info. 3 lower dimensions represent a single input, and an optional 4th dimension for batch of inputs. Data types supported: QASYMM8_SIGNED/QASYMM8/F16/F32.
- * @param[in] weights The 4d weights info with dimensions [width, height, IFM, OFM]. Data type supported: Same as @p input.
- * @param[in] bias (Optional) The biases have one dimension. Data type supported: Same as @p input.
+ * @param[in] weights The 4d weights info with dimensions [width, height, IFM, OFM]. Data type supported: Same as @p input or QSYMM8_PER_CHANNEL if @p input is QASYMM8/QASYMM8_SIGNED.
+ * @param[in] bias (Optional) The biases have one dimension. Data type supported: Should match @p input data type, except for input of QASYMM8 and QASYMM8_SIGNED type where biases should be of S32 type
* @param[in] output Output tensor info. The output has the same number of dimensions as the @p input.
* @param[in] deconv_info Contains padding and policies to be used in the deconvolution, this is described in @ref PadStrideInfo.
* @param[in] weights_info (Optional) Weights information needed for @ref CLConvolutionLayer, specifies if the weights tensor has been reshaped with @ref CLWeightsReshapeKernel.