perf(restrict3): shrink X-pass ii sweep to required overlap window

- compute fi_min/fi_max from output i-range and derive ii_lo/ii_hi
 - replace full ii sweep (-1:extf(1)) with windowed sweep in Z/Y precompute passes
 - keep stencil math unchanged; add bounds sanity check for ii window

AMSS_NCKU_source/prolongrestrict_cell.f90

@@ -2355,6 +2355,7 @@ end do
   real*8 :: tmp_xz_plane(-1:extf(1), 6)
   real*8 :: tmp_x_line(-1:extf(1))
   integer :: fi, fj, fk, ii, jj, kk
+  integer :: fi_min, fi_max, ii_lo, ii_hi
 
   if(wei.ne.3)then
      write(*,*)"prolongrestrict.f90::restrict3: this routine only surport 3 dimension"
@@ -2436,6 +2437,18 @@ end do
 
   call symmetry_bd(2,extf,funf,funff,SoA)
 
+  ! 仅计算 X 向最终写回所需的窗口：
+  ! func(i,j,k) 只访问 tmp_x_line(fi-2:fi+3)
+  fi_min = 2*(imino + lbc(1) - 1) - 1 - lbf(1) + 1
+  fi_max = 2*(imaxo + lbc(1) - 1) - 1 - lbf(1) + 1
+  ii_lo = fi_min - 2
+  ii_hi = fi_max + 3
+  if(ii_lo < -1 .or. ii_hi > extf(1))then
+      write(*,*)"restrict3: invalid ii window",ii_lo,ii_hi
+      write(*,*)"imino,imaxo,lbc(1),lbf(1),extf(1) = ",imino,imaxo,lbc(1),lbf(1),extf(1)
+      stop
+  endif
+
 !~~~~~~> restriction start...
 do k = kmino, kmaxo
     fk = 2*(k + lbc(3) - 1) - 1 - lbf(3) + 1
@@ -2446,7 +2459,7 @@ do k = kmino, kmaxo
         ! 优化点 1: 显式展开 Z 方向计算，减少循环开销
         ! 确保 ii 循环是最内层且连续访问
         !DIR$ VECTOR ALWAYS
-        do ii = -1, extf(1)
+        do ii = ii_lo, ii_hi
             ! 预计算当前 j 对应的 6 行在 Z 方向的压缩结果
             ! 这里直接硬编码 jj 的偏移，彻底消除一层循环
             tmp_xz_plane(ii, 1) = C1*(funff(ii,fj-2,fk-2)+funff(ii,fj-2,fk+3)) + &
@@ -2471,7 +2484,7 @@ do k = kmino, kmaxo
 
         ! 优化点 2: 同样向量化 Y 方向压缩
         !DIR$ VECTOR ALWAYS
-        do ii = -1, extf(1)
+        do ii = ii_lo, ii_hi
             tmp_x_line(ii) = C1*(tmp_xz_plane(ii, 1) + tmp_xz_plane(ii, 6)) + &
                             C2*(tmp_xz_plane(ii, 2) + tmp_xz_plane(ii, 5)) + &
                             C3*(tmp_xz_plane(ii, 3) + tmp_xz_plane(ii, 4))