COMPMID-2309 : CLConvolutionLayer: support QUANT8_SYMM_PER_CHANNEL filters
Change-Id: I16f6758b768ede404a064db057302ded706e1e8a
Signed-off-by: Vidhya Sudhan Loganathan <vidhyasudhan.loganathan@arm.com>
Signed-off-by: Michele Di Giorgio <michele.digiorgio@arm.com>
Reviewed-on: https://review.mlplatform.org/c/2215
Tested-by: Arm Jenkins <bsgcomp@arm.com>
Reviewed-by: Georgios Pinitas <georgios.pinitas@arm.com>
Reviewed-by: Gian Marco Iodice <gianmarco.iodice@arm.com>
Comments-Addressed: Arm Jenkins <bsgcomp@arm.com>
diff --git a/arm_compute/runtime/CL/functions/CLGEMMConvolutionLayer.h b/arm_compute/runtime/CL/functions/CLGEMMConvolutionLayer.h
index 017bf78..3392f11 100644
--- a/arm_compute/runtime/CL/functions/CLGEMMConvolutionLayer.h
+++ b/arm_compute/runtime/CL/functions/CLGEMMConvolutionLayer.h
@@ -60,7 +60,7 @@
/** Set the input and output tensors.
*
* @param[in] weights Weights tensor. Weights are 4D tensor with dimensions [kernel_x, kernel_y, IFM, OFM].
- * Data type supported: QASYMM8/F16/F32.
+ * Data type supported: QASYMM8/QSYMM8_PER_CHANNEL/F16/F32.
* @param[in] biases Biases tensor. Shared biases supported. Biases are 1D tensor with dimensions [OFM]. Data type supported: Same as @p weights.
* @param[out] output Destination tensor. Data types supported: Same as @p weights.
* @param[in] num_groups (Optional) Number of groups when performing a grouped convolution. num_groups != 1 is only supported for NCHW data layout
@@ -69,7 +69,7 @@
/** Static function to check if given info will lead to a valid configuration of @ref CLConvolutionLayerReshapeWeights
*
* @param[in] weights Weights tensor. Weights are 4D tensor with dimensions [kernel_x, kernel_y, IFM, OFM].
- * Data type supported: QASYMM8/F16/F32.
+ * Data type supported: QASYMM8/QSYMM8_PER_CHANNEL/F16/F32.
* @param[in] biases Biases tensor. Shared biases supported. Biases are 1D tensor with dimensions [OFM]. Data type supported: Same as @p weights.
* @param[in] output Destination tensor. Data types supported: Same as @p weights.
* @param[in] num_groups (Optional) Number of groups when performing a grouped convolution. num_groups != 1 is only supported for NCHW data layout
@@ -168,7 +168,8 @@
* @param[in] input Source tensor. 3 lower dimensions represent a single input [width, height, IFM],
* while every optional dimension from 4 and above represent a batch of inputs.
* Data types supported: QASYMM8/F16/F32.
- * @param[in] weights Weights tensor. Weights are 4D tensor with dimensions [kernel_x, kernel_y, IFM, OFM]. Data type supported: Same as @p input.
+ * @param[in] weights Weights tensor. Weights are 4D tensor with dimensions [kernel_x, kernel_y, IFM, OFM].
+ * Data type supported: Same as @p input or QASYMM8/QSYMM8_PER_CHANNEL when @p input is QASYMM8.
* @param[in] biases Biases tensor. Shared biases supported. Biases are 1D tensor with dimensions [OFM].
* Data type supported: Should match @p input data type, except for input of QASYMM8 type where biases should be of S32 type.
* @param[out] output Destination tensor. 3 lower dimensions represent a single output [width, height, OFM], while the rest represent batch of outputs.
@@ -187,7 +188,8 @@
* @param[in] input Source tensor. 3 lower dimensions represent a single input [width, height, IFM],
* while every optional dimension from 4 and above represent a batch of inputs.
* Data types supported: QASYMM8/F16/F32.
- * @param[in] weights Weights tensor. Weights are 4D tensor with dimensions [kernel_x, kernel_y, IFM, OFM]. Data type supported: Same as @p input.
+ * @param[in] weights Weights tensor. Weights are 4D tensor with dimensions [kernel_x, kernel_y, IFM, OFM].
+ * Data type supported: Same as @p input or QASYMM8/QSYMM8_PER_CHANNEL when @p input is QASYMM8.
* @param[in] biases Biases tensor. Shared biases supported. Biases are 1D tensor with dimensions [OFM].
* Data type supported: Should match @p input data type, except for input of QASYMM8 type where biases should be of S32 type.
* @param[out] output Destination tensor. 3 lower dimensions represent a single output [width, height, OFM], while the rest represent batch of outputs.
@@ -212,7 +214,7 @@
/** Configures the appropriate matrix multiply routine
*
* @param[in] input Input tensor. Data types supported: QASYMM8/F16/F32.
- * @param[in] weights Weights tensor. Data type supported: Same as @p input.
+ * @param[in] weights Weights tensor. Data type supported: Same as @p input or QASYMM8/QSYMM8_PER_CHANNEL when @p input is QASYMM8.
* @param[in] biases Biases tensor. Shared biases supported. Biases are 1D tensor with dimensions [OFM].
* Data type supported: Should match @p input data type, except for input of QASYMM8 type where biases should be of S32 type.
* @param[in, out] output Output tensor. Data types supported: Same as @p input,
@@ -225,12 +227,12 @@
const ActivationLayerInfo &act_info);
/** Static function to check if given info will lead to a valid configuration of @ref CLGEMMConvolutionLayer matrix multiply routines
*
- * @param[in] input Input tensor. Data types supported: QASYMM8/F16/F32.
- * @param[in] weights Weights tensor. Data type supported: Same as @p input.
- * @param[in] output Output tensor. Data types supported: Same as @p input,
- * except for input of QASYMM8 type where output should be of S32 type.
- * @param[in] biases Biases tensor. Shared biases supported. Biases are 1D tensor with dimensions [OFM].
+ * @param[in] input Input tensor info. Data types supported: QASYMM8/F16/F32.
+ * @param[in] weights Weights tensor info. Data type supported: Same as @p input or QASYMM8/QSYMM8_PER_CHANNEL when @p input is QASYMM8.
+ * @param[in] biases Biases tensor info. Shared biases supported. Biases are 1D tensor with dimensions [OFM].
* Data type supported: Should match @p input data type, except for input of QASYMM8 type where biases should be of S32 type.
+ * @param[in] output Output tensor info. Data types supported: Same as @p input,
+ * except for input of QASYMM8 type where output should be of S32 type.
* @param[in] gemmlowp_output_stage GEMMLowp output stage info
* @param[in] gemm_3d_depth Depth of GEMM 3D
* @param[in] skip_im2col Flag which specifies if im2col has to be skipped. i.e. 1x1 convolution with NHWC data layout.