INT8 Quantized MatMul

题面

Implement INT8 量化矩阵乘法：给定 $A \in \mathbb{Z}_8^{M \times K}$、$B \in \mathbb{Z}_8^{K \times N}$，量化尺度 $s_A, s_B, s_C$ 与零点 $z_A, z_B, z_C$，按如下公式计算输出 $C_{quant}$（int8）：

$$C_{quant}(i,j) = \text{clamp}\left(\text{round}\left( \frac{s_A s_B}{s_C} \sum_k (A_{ik} - z_A)(B_{kj} - z_B) \right) + z_C, -128, 127 \right)$$

计算输出 C（int8）。

Implementation Requirements

• External libraries are not permitted
• The solve function signature must remain unchanged
• Accumulate in int32, scale in float32，round 最近整数，加零点并 clamp 至 [-128,127]
• The final result must be stored in int8 矩阵 C

Examples

Example 1: 见页面示例（M=N=K=2）
Example 2: 含非零零点示例

Constraints

• 1 ≤ M, N, K ≤ 4096
• scale_* 为正浮点数
• -128 ≤ zero_point_* ≤ 127
• Performance is measured with K = 2,048, M = 8,192, N = 4,096