diff options
author | Georgios Pinitas <georgios.pinitas@arm.com> | 2017-07-12 12:30:40 +0100 |
---|---|---|
committer | Anthony Barbier <anthony.barbier@arm.com> | 2018-09-17 14:16:42 +0100 |
commit | e222941f931b1d44bb29e5827b6df748e60cefc4 (patch) | |
tree | 1a54a60750caa998527e5c6b2a494047cb3ccc23 /arm_compute/runtime/NEON | |
parent | bbd3d6045ba6480be350c0b1610048ee6c9c050f (diff) | |
download | ComputeLibrary-e222941f931b1d44bb29e5827b6df748e60cefc4.tar.gz |
COMPMID-401: Implement FixedPointPosition conversion for NEON.
Adds support of changing the fixed point position of a tensor in
DepthConvert.
Change-Id: Ic3b50a4628fac7497a0217d92941c9d6f64d21cb
Reviewed-on: http://mpd-gerrit.cambridge.arm.com/80438
Reviewed-by: Anthony Barbier <anthony.barbier@arm.com>
Tested-by: Kaizen <jeremy.johnson+kaizengerrit@arm.com>
Diffstat (limited to 'arm_compute/runtime/NEON')
-rw-r--r-- | arm_compute/runtime/NEON/functions/NEDepthConvert.h | 21 |
1 files changed, 11 insertions, 10 deletions
diff --git a/arm_compute/runtime/NEON/functions/NEDepthConvert.h b/arm_compute/runtime/NEON/functions/NEDepthConvert.h index 47b3a7e6f6..37f7293fb3 100644 --- a/arm_compute/runtime/NEON/functions/NEDepthConvert.h +++ b/arm_compute/runtime/NEON/functions/NEDepthConvert.h @@ -45,24 +45,25 @@ public: const NEDepthConvert &operator=(const NEDepthConvert &) = delete; /** Initialize the function's source, destination * - * Input format must be different than output format. - * * Valid conversions Input -> Output : - * QS8 -> F32 + * QS8 -> QS8, F32 * U8 -> U16, S16, S32 * U16 -> U8, U32 * S16 -> U8, S32 - * QS16 -> F32 + * QS16 -> QS16, F32 * F32 -> QS8, QS16 * + * @warning In case of in-place fixed point position conversion make sure that configure has been called + * before the updated tensor is used in other functions, as the TensorInfo of the tensor will be + * altered. In-place is only supported for QS8 -> QS8, QS16 -> QS16. * - * @param[in] input The input tensor to convert. Data type supported: QS8/U8/U16/S16/QS16/F32. - * @param[out] output The output tensor. Data type supported: QS8/U8/U16/S16/QS16/U32/S32/F32. - * @param[in] policy Conversion policy. - * @param[in] shift Value for down/up conversions. Must be 0 <= shift < 8. - * It is not used on fixed point conversion. + * @param[in, out] input The input tensor to convert (Written in case of in-place computation). Data types supported: U8/QS8/U16/S16/F32. + * @param[out] output The output tensor. Can be null in case of in-place computation. Data types supported: U8/QS8/U16/S16/U32/S32/F32. + * @param[in] policy Conversion policy. + * @param[in] shift (Optional) Value for down/up conversions. Must be 0 <= shift < 8. + * In case of fixed point position conversion, it specifies the new fixed point position, if operation is in-place. */ - void configure(const ITensor *input, ITensor *output, ConvertPolicy policy, uint32_t shift); + void configure(ITensor *input, ITensor *output, ConvertPolicy policy, uint32_t shift = 0); }; } #endif /*__ARM_COMPUTE_NEDEPTHCONVERT_H__*/ |