Fortran77代码触发Segmentation Fault（内存访问问题）求助

终端报错信息 #

Program received signal SIGSEGV, Segmentation fault.
0x0000000000f523b7 in bcvnmtn (xb=<error reading variable: Cannot access memory at address 0xc44a8000c3818>, 
    yb=<error reading variable: Cannot access memory at address 0xc5dc8000c5138>, nstp=0, npoints=0, tx=..., ty=..., anx=..., any=...) at ./bcvnmtn.F:14
14            tx(1) = xb(1+nstp) - xb(1)

触发段错误的子例程 bcvnmtn #

subroutine bcvnmtn(xb,yb,nstp,npoints,tx,ty,anx,any)

c-----compute the vertex valued tangential and normal vector fields
c----- (tx,ty)(n) and (anx,any)(n) to the curve (xb,yb)((n-1)*nstp+1)
c----- for n=1 to npoints

cdir$ nolist
      include 'paramcom.h'
cdir$ list
      dimension xb(1),yb(1),tx(1),ty(1),anx(1),any(1)

c-----get tangent vector field

      tx(1) = xb(1+nstp) - xb(1)
      ty(1) = yb(1+nstp) - yb(1)

      in = 1
      do 100 n=2,npoints-1
        inp = in + 2*nstp
        tx(n) = (xb(inp) - xb(in))*0.5
        ty(n) = (yb(inp) - yb(in))*0.5
        in = in + nstp
  100 continue

      nlast = npoints
      ilast = 1 + (npoints-1)*nstp
      tx(nlast) = xb(ilast) - xb(ilast-nstp)
      ty(nlast) = yb(ilast) - yb(ilast-nstp)

c-----normalize to unit length and rotate to get normal vector

      do 200 n=1,npoints
        tl = sqrt( tx(n)**2 + ty(n)**2 )
        tx(n) = tx(n) / ( tl + tiny )
        ty(n) = ty(n) / ( tl + tiny )
        any(n) = tx(n)
        anx(n) = -ty(n)
  200 continue

      return
      end

调用 bcvnmtn 的子例程 bcsymgd #

subroutine bcsymgd(xi,yi)

c-----compute ghost point locations along a symmetry boundary

cdir$ nolist
      include 'common.h'
      include 'inputcom.h'
      include 'pointer.h'
cdir$ list
      common /bcpoint/ idimfrm,jdimfrm,isttfrm,jsttfrm,nstpfrm,nofffrm,
     %                 idimtoo,jdimtoo,istttoo,jstttoo,nstptoo,nofftoo,
     %                 npoints
      dimension xi(0:idimfrm,0:jdimfrm),yi(0:idimfrm,0:jdimfrm)
      call bcvnmtn(xi(isttfrm,jsttfrm),yi(isttfrm,jsttfrm),
     %             nstpfrm,npoints,tangx,tangy,anormx,anormy)
      call bcdifvf(dispx(isttfrm,jsttfrm),dispy(isttfrm,jsttfrm),
     %             nofffrm,nstpfrm,
     %             xi(isttfrm,jsttfrm),yi(isttfrm,jsttfrm),
     %             (- nofftoo),nstpfrm,
     %             xi(isttfrm,jsttfrm),yi(isttfrm,jsttfrm),
     %             nofffrm,nstpfrm,npoints)
      call bctrflc(dispx(isttfrm,jsttfrm),dispy(isttfrm,jsttfrm),
     %             dispx(isttfrm,jsttfrm),dispy(isttfrm,jsttfrm),
     %             nstpfrm,npoints,tangx,tangy,anormx,anormy)
      call bcaddvf(xi(istttoo,jstttoo),yi(istttoo,jstttoo),
     %             nofftoo,nstptoo,
     %             dispx(isttfrm,jsttfrm),dispy(isttfrm,jsttfrm),
     %             nofffrm,nstpfrm,
     %             xi(isttfrm,jsttfrm),yi(isttfrm,jsttfrm),
     %             nofffrm,nstpfrm,npoints)

      return
      end

问题背景 #

• bcsymgd 中的 xi 和 yi 为二维几何坐标数组，调用 bcvnmtn 时转换为一维数组 xb、yb 传入
• 使用编译器：pgi/2020.4
• 代码为老旧大型程序，修改权限有限

错误原因分析 #

从报错信息可见核心异常：

1. nstp=0、npoints=0，直接导致子例程内数组访问逻辑完全失效，例如第14行的xb(1+nstp)虽未越界，但后续循环和末尾的数组索引计算会出现非法内存访问
2. xb 和 yb 无法读取内存，说明传入的数组指针无效，大概率是isttfrm、jsttfrm超出xi、yi的数组范围，或nstpfrm、npoints未正确初始化

排查建议 #

1. 检查公共块参数初始化：确认/bcpoint/公共块中的nstpfrm、npoints、isttfrm、jsttfrm在调用bcsymgd前已正确赋值，尤其npoints和nstpfrm不能为0
2. 验证数组边界：检查xi(isttfrm,jsttfrm)的索引是否在xi的有效范围（0:idimfrm, 0:jdimfrm）内，索引越界会传入无效指针
3. 开启编译边界检查：用PGI编译器的-C（数组边界检查）和-g（调试信息）选项重新编译，获取更详细的越界位置：

   pgf77 -g -C your_program.f -o your_program
   

4. 调试跟踪参数值：用GDB在bcsymgd入口设断点，查看公共块参数实际值：

   gdb ./your_program
   break bcsymgd
   run
   print idimfrm, jdimfrm, isttfrm, jsttfrm, nstpfrm, npoints
   

5. 检查输出数组维度：确认tangx、tangy、anormx、anormy的维度至少为npoints，避免写入越界

内容的提问来源于stack exchange，提问作者Surya Sarvajith