FP32 Tensor to INT8 Quantization (with Asymmetric Quantization Follow-ups)

Given an FP32 tensor x (treat it as a 1-D array / contiguous buffer) and a floating-point scale, quantize it into an INT8 tensor q. Base version (symmetric) For each element: q[i] = (int8)( x[i] / scale ) Follow-up (asymmetric quantization) Extend the base version to support asymmetric quantization with an additional integer parameter zero_point: q[i] = (int8)( x[i] / scale + zero_point ) Requirements Provide/implement a function interface (C++ or pseudocode) to perform the conversion. Discuss why zero_point is needed for non-symmetric input distributions, and how scale + zero_point work together. Sample tests x=[0.0, 1.0, -1.0], scale=0.5, zero_point=0 → q=[0, 2, -2] x=[0.0, 1.0, -1.0], scale=0.5, zero_point=10 → q=[10, 12, 8] x=[2.5], scale=0.5, zero_point=0 → q=[5] x=[-2.5], scale=0.5, zero_point=0 → q=[-5] x=[0.2, 0.3], scale=0.1, zero_point=0 → q=[2, 3] Constraints Let n be the number of elements. 1 ≤ n ≤ 10^7 (implementation should be O(n) with a single pass). Example Input 3 0.0 1.0 -1.0 0.5 0 Output 0 2 -2