diff options
| author | Georgios Pinitas <georgios.pinitas@arm.com> | 2020-03-06 18:12:09 +0000 |
|---|---|---|
| committer | Georgios Pinitas <georgios.pinitas@arm.com> | 2020-03-12 12:12:30 +0000 |
| commit | c7b183ab741650653289f8ce3bdeb4926521fdbd (patch) | |
| tree | 991e9f20340c91c288d52d8f9a64a3729e4a40b0 /arm_compute/core | |
| parent | 6800117df3be825f0ec5c6cc71c4377322f51b99 (diff) | |
| download | ComputeLibrary-c7b183ab741650653289f8ce3bdeb4926521fdbd.tar.gz | |
COMPMID-3160: Add Bfloat16 support in NEGEMMConvolutionLayer
Signed-off-by: Georgios Pinitas <georgios.pinitas@arm.com>
Change-Id: I0e449306c138a562ffc1455e76ec44b2fd059d85
Reviewed-on: https://review.mlplatform.org/c/ml/ComputeLibrary/+/2860
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 'arm_compute/core')
| -rw-r--r-- | arm_compute/core/NEON/kernels/NEIm2ColKernel.h | 6 | ||||
| -rw-r--r-- | arm_compute/core/NEON/kernels/NEWeightsReshapeKernel.h | 6 | ||||
| -rw-r--r-- | arm_compute/core/Utils.h | 2 |
3 files changed, 8 insertions, 6 deletions
diff --git a/arm_compute/core/NEON/kernels/NEIm2ColKernel.h b/arm_compute/core/NEON/kernels/NEIm2ColKernel.h index 292e941272..1c358b379d 100644 --- a/arm_compute/core/NEON/kernels/NEIm2ColKernel.h +++ b/arm_compute/core/NEON/kernels/NEIm2ColKernel.h @@ -77,7 +77,8 @@ public: /** Set the input and output of the kernel. * * @param[in] input The input tensor to convert. 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/QASYMM8_SIGNED/F16/F32 + * while every optional dimension from 4 and above represent a batch of inputs. + * Data types supported: QASYMM8/QASYMM8_SIGNED/BFLOAT16/F16/F32 * Note: QASYMM8 works only for has_bias = false * @param[out] output The output tensor. Data types supported: Same as @p input * @param[in] kernel_dims The kernel dimensions (width and height). @@ -91,7 +92,8 @@ public: /** Static function to check if given info will lead to a valid configuration of @ref NEIm2ColKernel * * @param[in] input The input tensor to convert. 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/QASYMM8_SIGNED/F16/F32 + * while every optional dimension from 4 and above represent a batch of inputs. + * Data types supported: QASYMM8/QASYMM8_SIGNED/BFLOAT16/F16/F32 * Note: QASYMM8 works only for has_bias = false * @param[in] output The output tensor. Data types supported: Same as @p input * @param[in] kernel_dims The kernel dimensions (width and height). diff --git a/arm_compute/core/NEON/kernels/NEWeightsReshapeKernel.h b/arm_compute/core/NEON/kernels/NEWeightsReshapeKernel.h index 46be5233eb..b68cb50c7b 100644 --- a/arm_compute/core/NEON/kernels/NEWeightsReshapeKernel.h +++ b/arm_compute/core/NEON/kernels/NEWeightsReshapeKernel.h @@ -1,5 +1,5 @@ /* - * Copyright (c) 2017-2019 ARM Limited. + * Copyright (c) 2017-2020 ARM Limited. * * SPDX-License-Identifier: MIT * @@ -76,7 +76,7 @@ public: * * @param[in] input The input tensor to convert. Weights are 4D tensor with dimensions [kernel_x, kernel_y, IFM, OFM] if shared, * and 5D tensor with dimensions [kernel_x, kernel_y, IFM, OFM, num_patches] if unshared. - * Data types supported: QASYMM8/QASYMM8_SIGNED/QSYMM8_PER_CHANNEL/FP16/F32 + * Data types supported: QASYMM8/QASYMM8_SIGNED/QSYMM8_PER_CHANNEL/BFLOAT16/FP16/F32 * @param[in] bias The shared biases tensor to append. Bias is 1D tensor with dimensions [OFM] if shared and 2D tensor with * dimensions [OFM, num_patches] if unshared. Data types supported: Same as @p input * @warning Appending biases to weights reshaped matrix is not supported for quantized asymmetric types. @@ -87,7 +87,7 @@ public: * * @param[in] input The input tensor to convert. Weights are 4D tensor with dimensions [kernel_x, kernel_y, IFM, OFM] if shared, * and 5D tensor with dimensions [kernel_x, kernel_y, IFM, OFM, num_patches] if unshared. - * Data types supported: QASYMM8/QASYMM8_SIGNED/QSYMM8_PER_CHANNEL/F16/F32 + * Data types supported: QASYMM8/QASYMM8_SIGNED/QSYMM8_PER_CHANNEL/BFLOAT16/F16/F32 * @param[in] biases The shared biases tensor to append. Bias is 1D tensor with dimensions [OFM] if shared and 2D tensor with * dimensions [OFM, num_patches] if unshared. Data types supported: Same as @p input * @warning Appending biases to weights reshaped matrix is not supported for quantized asymmetric types. diff --git a/arm_compute/core/Utils.h b/arm_compute/core/Utils.h index 8577046af0..eff6157b1f 100644 --- a/arm_compute/core/Utils.h +++ b/arm_compute/core/Utils.h @@ -1342,7 +1342,7 @@ void print_consecutive_elements_impl(std::ostream &s, const T *ptr, unsigned int } else if(std::is_same<typename std::decay<T>::type, bfloat16>::value) { - // We use T instead of print_type here is because the std::is_floating_point<bfloat> returns false and then the print_type becomes int. + // We use T instead of print_type here is because the std::is_floating_point<bfloat16> returns false and then the print_type becomes int. s << std::right << float(ptr[i]) << element_delim; } else |