COMPMID-2418: CLDequantizationLayer support for QASYMM8_PER_CHANNEL
Add support for QASYMM8_PER_CHANNEL in CLDequantiazationLayer.
Added tests for NHWC and also updated NEON code to work with NHWC
data layout.
Cleaned up the reference implementation.
Change-Id: Ic1d51f16f7f625503fffdbbb66f6487aa588f08c
Signed-off-by: Michalis Spyrou <michalis.spyrou@arm.com>
Reviewed-on: https://review.mlplatform.org/c/1828
Tested-by: Arm Jenkins <bsgcomp@arm.com>
Comments-Addressed: Arm Jenkins <bsgcomp@arm.com>
Reviewed-by: Georgios Pinitas <georgios.pinitas@arm.com>
diff --git a/arm_compute/runtime/CL/CLTensorAllocator.h b/arm_compute/runtime/CL/CLTensorAllocator.h
index 982cc51..f7800d3 100644
--- a/arm_compute/runtime/CL/CLTensorAllocator.h
+++ b/arm_compute/runtime/CL/CLTensorAllocator.h
@@ -146,8 +146,8 @@
CLMemory _memory; /**< OpenCL memory */
uint8_t *_mapping; /**< Pointer to the CPU mapping of the OpenCL buffer. */
CLTensor *_owner; /**< Owner of the allocator */
- CLFloatArray _scale;
- CLInt32Array _offset;
+ CLFloatArray _scale; /**< Scales array in case of quantized per channel data type */
+ CLInt32Array _offset; /**< Offsets array in case of quantized per channel data type */
};
} // namespace arm_compute
#endif /* __ARM_COMPUTE_CLTENSORALLOCATOR_H__ */