Optimize vloc_of_g by skipping negligible analytic terms

[A-006]

Summary·

This PR introduces a new utility header source/source_base/truncated_func.h to encapsulate truncated math functions (truncated_exp, truncated_erfc) and architecture-specific underflow check optimizations (truncated_underflow). This refactoring improves code readability, reduces duplication of __SW specific bitwise logic in kernel operations, and centralizes the handling of numerical underflow/cutoff.

Modifications

1. New Feature: source/source_base/truncated_func.h

   • Implemented ModuleBase::truncated_exp: Exponential function with underflow protection.
   • Implemented ModuleBase::truncated_erfc: Complementary error function with cutoff for large arguments.
   • Implemented ModuleBase::truncated_underflow: Optimized function to force small values to zero. It utilizes IEEE 754 bitwise operations for double and float types to efficiently detect and handle underflow, replacing verbose inline code primarily used for __SW (Sunway) architecture optimizations.

2. Refactoring:

   • source/source_pw/module_pwdft/kernels/force_op.cpp: Simplified __SW macro blocks by replacing manual bitwise exponent checks with ModuleBase::truncated_underflow.
   • source/source_pw/module_pwdft/kernels/stress_op.cpp: Applied ModuleBase::truncated_underflow for optimization and adopted ModuleBase::truncated_exp.
   • source/source_pw/module_pwdft/forces.cpp: Updated references to use the new ModuleBase::truncated_erfc and ModuleBase::truncated_exp.
   • source/source_pw/module_pwdft/vl_pw.cpp: Updated references to use ModuleBase::truncated_exp.