0

I am implementing event detection in an ODE solver suite. Following is a piece of code in the implementation that I am currently debugging and notice a strange behavior. Once the root finding routine returns successfully with a root, the first if block is executed (ROOTFINDING_ERR == 0) -- lines 1 to 14 -- which is expected behavior. However, once the block finishes executing, execution jumps unexpectedly to the last statement of the ELSE block corresponding to (ROOTFINDING_ERR == 0) -- line 20 -- and the statement EVENT_OUT(j) = .FALSE. is executed before executing ENDIF on line 21.

               1  IF (ROOTFINDING_ERR == 0) THEN
               2     IF ((ABS(EVENT_TIMES(1, j) - T_STAR) > &
               3          NEARBY_ROOTS_ABSTOL)  &
               4          .AND. (EVENT_ITER < EVENT_ITER_MAX)) THEN
               5        EVENT_TIMES(1, j) = T_STAR
               6        EVENT_TIMES(2, j) = DBLE(j)
               7        EVENT_TEST(j)     = .TRUE.
               8        EVENT_OUT(j)      = .TRUE.
               9     ELSE
              10        EVENT_TIMES(1, j) = T_STAR
              11        EVENT_TIMES(2, j) = DBLE(j)
              12        EVENT_TEST(j) = .FALSE.
              13        EVENT_OUT(j)  = .TRUE.
              14     ENDIF
              15  ELSE
              16     EVENT_TIMES(1, j) = T_STAR
              17     EVENT_TIMES(2, j) = DBLE(j)
              18     ! Dont test this event further
              19     EVENT_TEST(j)     = .FALSE.
              20     EVENT_OUT(j)      = .FALSE.
              21  ENDIF

I assumed this has to do with some memory access/allocation problems and ran DrMemory (a tool like Valgrind) on the executable file. DrMemory did specify that some reads are uninitialized which I will look further into.

I was wondering where to go from here. I did step through the code using gdb and had some difficulty isolating "uninitialized read" that DrMemory complained about. I turned on -Wall, -Wextra, -fbounds-check etc to figure out if the compiler is able to catch it. I also tried to disassemble the exe with x32dbg, but the debug symbols are not read as x32dbg doesn't support DWARF symbols.

4

1 回答 1

1

From the code shown I see no reason for the error at all. I surmise therefore that the error is caused by something we cannot see.

Now, what follows doesn't solve the puzzle, but seeks to go round it ...

... the code shown could be simplified to 

EVENT_TIMES(1, j) = T_STAR
EVENT_TIMES(2, j) = DBLE(j)
IF (ROOTFINDING_ERR == 0) THEN
   EVENT_OUT(j)      = .TRUE.
   IF ((ABS(EVENT_TIMES(1, j) - T_STAR) > &
        NEARBY_ROOTS_ABSTOL)  &
        .AND. (EVENT_ITER < EVENT_ITER_MAX)) THEN
      EVENT_TEST(j)     = .TRUE.
   ELSE
      EVENT_TEST(j) = .FALSE.
   ENDIF
ELSE
   ! Dont test this event further
   EVENT_TEST(j)     = .FALSE.
   EVENT_OUT(j)      = .FALSE.
ENDIF

... or even further to ...

EVENT_TIMES(1, j) = T_STAR
EVENT_TIMES(2, j) = DBLE(j)
EVENT_OUT(j) = ROOTFINDING_ERR == 0
EVENT_TEST(j)     = ((ROOTFINDING_ERR == 0) .AND. &
                    (ABS(EVENT_TIMES(1, j) - T_STAR) > &
                    NEARBY_ROOTS_ABSTOL)  &
                    .AND. (EVENT_ITER < EVENT_ITER_MAX))
于 2018-01-04T10:53:35.623 回答