• Previous message: Jonathan B. Liu: "a way to fix magic color problem"
• Messages sorted by: [ date ] [ thread ] [ subject ] [ author ]

performance bug:
 loading a file containing even a modest number (5000) of subcell calls
with invalid timestamps takes > 2 hours on PII/300MHz
after the patch:  ~10 seconds.

the problem is that magic was repeatedly loading and computing the
bounding box of each subcell, the fix just sets a flag to indicate this
computation is done.

patch to database/database.h  database/DBtimestmp.c
this also fixes a use after free memory allocation bug.

--conrad




*** magic-7.1.5/database/database.h	Wed Feb 27 11:22:26 2002
--- local/database/database.h	Sun May 12 18:00:20 2002
***************
*** 288,293 ****
--- 288,295 ----
  #define CDNOEDIT	0200
  #endif

+ #define CDPROCESSED    0400
+
  /*
   * Description of an array.
   * The bounds xlo .. xhi and ylo .. yhi are transformed versions

*** magic-7.1.5/database/DBtimestmp.c	Wed Nov 28 14:23:34 2001
--- local/database/DBtimestmp.c	Sun May 12 18:08:06 2002
***************
*** 109,115 ****
      Rect oldArea, parentArea, tmp;
      int redisplay;
      int firstOne = TRUE;
!
      /* It's very important to disable interrupts during this section!
       * Otherwise, we may not paint the recheck tiles properly.
       */
--- 109,116 ----
      Rect oldArea, parentArea, tmp;
      int redisplay;
      int firstOne = TRUE;
!     void *tofree;
!     Mismatch *tmpm;
      /* It's very important to disable interrupts during this section!
       * Otherwise, we may not paint the recheck tiles properly.
       */
***************
*** 118,123 ****
--- 119,129 ----
      if (mismatch == NULL) return;
      TxPrintf("Processing timestamp mismatches:");
      SigDisableInterrupts();
+     for(tmpm=mismatch;tmpm!=NULL;tmpm=tmpm->mm_next)
+       {
+ 	tmpm->mm_cellDef->cd_flags&=(~CDPROCESSED);
+       }
+
      while (mismatch != NULL)
      {
  	/* Be careful to remove the front element from the mismatch
***************
*** 127,134 ****

  	cellDef = mismatch->mm_cellDef;
  	oldArea = mismatch->mm_oldArea;
! 	FREE((char *) mismatch);
  	mismatch = mismatch->mm_next;
  	(void) DBCellRead(cellDef, (char *) NULL, TRUE);

  	/* Jimmy up the cell's current bounding box, so the following
--- 133,144 ----

  	cellDef = mismatch->mm_cellDef;
  	oldArea = mismatch->mm_oldArea;
! 	tofree=mismatch;
  	mismatch = mismatch->mm_next;
+ 	FREE((char *) tofree);
+
+ 	if( ((cellDef->cd_flags)&(CDPROCESSED))==0)
+ 	  {
  	    (void) DBCellRead(cellDef, (char *) NULL, TRUE);

  	    /* Jimmy up the cell's current bounding box, so the following
***************
*** 168,174 ****
--- 178,187 ----
  	      }
  	    else TxPrintf(", %s", cellDef->cd_name);
  	    TxFlush();	/* This is needed to prevent _doprnt screwups */
+ 	    cellDef->cd_flags|=CDPROCESSED;
+ 	  }
      }
+
      SigEnableInterrupts();
      TxPrintf(".\n");
      TxFlush();
***************
*** 274,279 ****
--- 287,293 ----
      mm->mm_oldArea = *wrongArea;
      mm->mm_next = mismatch;
      mismatch = mm;
+     cellDef->cd_flags&=(~CDPROCESSED);

      for (parentUse = cellDef->cd_parents; parentUse != NULL;
  	parentUse = parentUse->cu_nextuse)

• Next message: R. Timothy Edwards: "magic patch"
• Previous message: Jonathan B. Liu: "a way to fix magic color problem"
• Messages sorted by: [ date ] [ thread ] [ subject ] [ author ]