CANN/mat-chem-sim-pred SOPDT批处理滚动评分API

PidSopdtBatchRolloutScore API

【免费下载链接】mat-chem-sim-pred面向工业领域，聚焦计算仿真、预测两大核心场景，构建面向流程工业"机理+数据"双轮驱动的领域计算层，推动AI for Science在材料化学领域的深度应用。项目地址: https://gitcode.com/cann/mat-chem-sim-pred

extern "C" uint64_t aclnnPidSopdtBatchRolloutScoreGetWorkspaceSize( int64_t batch, int64_t candidates, int64_t sim_steps, int64_t candidate_tile); extern "C" int32_t aclnnPidSopdtBatchRolloutScore( void* a1, void* a2, void* b, void* delay, void* y0, void* sp, void* kp, void* ki, void* kd, void* best_result, void* best_idx, int64_t batch, int64_t candidates, int64_t sim_steps, int64_t candidate_tile, float sample_interval, float settle_band, float overshoot_weight, float settling_weight, float control_weight, void* workspace, uint64_t workspace_size, void* stream);

Output Layout

best_score,best_kp,best_ki,best_kd,best_iae,best_ise,best_overshoot,best_settling_time

Notes

• SOPDT takes two output-history coefficients: the plant isy[k+1] = a1*y[k] + a2*y[k-1] + b*u[k-delay]
• candidate_tilecontrols how many PID candidates are evaluated in one local-kernel launch. Pass0(the default) to let the host auto-select the optimal tilemin(candidates, kLane=768);768is the vector-lane throughput sweet spot, so callers normally leave it at0.
• delayis clamped to0..31inside the kernel
• workspace stores tile-local best results and the final reduction metadata

【免费下载链接】mat-chem-sim-pred面向工业领域，聚焦计算仿真、预测两大核心场景，构建面向流程工业"机理+数据"双轮驱动的领域计算层，推动AI for Science在材料化学领域的深度应用。项目地址: https://gitcode.com/cann/mat-chem-sim-pred