[Teem-commit] teem/src/hoover hoover.h,1.23,1.24 methodsHoover.c,1.10,1.11 rays.c,1.30,1.31

[Gordon K. <kin...@us...>]

Update of /cvsroot/teem/teem/src/hoover
In directory sc8-pr-cvs1.sourceforge.net:/tmp/cvs-serv4697

Modified Files:
	hoover.h methodsHoover.c rays.c 
Log Message:
improved load balancing of hooverRender- now scanlines are doled out as fast as they are finished, rather than assigning each thread to every Nth scanline

Index: hoover.h
===================================================================
RCS file: /cvsroot/teem/teem/src/hoover/hoover.h,v
retrieving revision 1.23
retrieving revision 1.24
diff -C2 -d -r1.23 -r1.24
*** hoover.h	19 Feb 2004 03:44:57 -0000	1.23
--- hoover.h	11 May 2004 06:00:17 -0000	1.24
***************
*** 84,88 ****
  ** 4) image information
  ** 5) opaque "user information" pointer
! ** 6) the number of threads to spawn
  ** 7) the callbacks
  */
--- 84,88 ----
  ** 4) image information
  ** 5) opaque "user information" pointer
! ** 6) stuff about multi-threading
  ** 7) the callbacks
  */
***************
*** 90,109 ****
  
    /******** 1) camera information */
!   limnCamera *cam;         /* camera info */
  
    /******** 2) volume information: size and spacing, centering */
!   int volSize[3];          /* X,Y,Z resolution of volume */
!   double volSpacing[3];    /* distance between samples in X,Y,Z direction */
!   int volCentering;        /* either nrrdCenterNode or nrrdCenterCell */
    
    /******** 3) image information: dimensions + centering */
!   int imgSize[2],          /* # samples of image along U and V axes */
!     imgCentering;          /* either nrrdCenterNode or nrrdCenterCell */
    
    /******** 4) opaque "user information" pointer */
!   void *user;          /* passed to all callbacks */
  
!   /******** 5) the number of threads to spawn */
!   int numThreads;          /* number of threads to spawn per rendering */
    
    /*
--- 90,111 ----
  
    /******** 1) camera information */
!   limnCamera *cam;           /* camera info */
  
    /******** 2) volume information: size and spacing, centering */
!   int volSize[3];            /* X,Y,Z resolution of volume */
!   double volSpacing[3];      /* distance between samples in X,Y,Z direction */
!   int volCentering;          /* either nrrdCenterNode or nrrdCenterCell */
    
    /******** 3) image information: dimensions + centering */
!   int imgSize[2],            /* # samples of image along U and V axes */
!     imgCentering;            /* either nrrdCenterNode or nrrdCenterCell */
    
    /******** 4) opaque "user information" pointer */
!   void *user;                /* passed to all callbacks */
  
!   /******** 5) stuff about multi-threading */
!   int numThreads,            /* number of threads to spawn per rendering */
!     workIdx;                 /* next work assignment (such as a scanline) */
!   airThreadMutex *workMutex; /* mutex around work assignment */
    
    /*
***************
*** 172,182 ****
    **
    ** This is not a terribly flexible scheme (don't forget, this is
!   ** hoover) in that it enforces rather rigid constraints on how
!   ** multi-threading works: one thread can not render multiple rays
    ** simulatenously.  If there were more args to sample() (like a
    ** ray, or an integral rayIndex), then this would be possible,
    ** but it would mean that _hooverThreadBody() would have to
!   ** implement all the smarts about which samples belong on which rays
!   ** belong with which threads.
    **
    ** At some point now or in the future, an effort will be made to
--- 174,184 ----
    **
    ** This is not a terribly flexible scheme (don't forget, this is
!   ** hoover) in that it imposes some constraints on how multi-threading
!   ** can work: one thread can not render multiple rays
    ** simulatenously.  If there were more args to sample() (like a
    ** ray, or an integral rayIndex), then this would be possible,
    ** but it would mean that _hooverThreadBody() would have to
!   ** implement all the smarts about which samples belong on which rays,
!   ** and which rays belong with which threads.
    **
    ** At some point now or in the future, an effort will be made to

Index: rays.c
===================================================================
RCS file: /cvsroot/teem/teem/src/hoover/rays.c,v
retrieving revision 1.30
retrieving revision 1.31
diff -C2 -d -r1.30 -r1.31
*** rays.c	10 May 2004 19:29:22 -0000	1.30
--- rays.c	11 May 2004 06:00:17 -0000	1.31
***************
*** 199,207 ****
    }
  
!   /* for now, the load-balancing among P processors is simplistic: the
!      Nth thread (0-based numbering) gets scanlines N, N+P, N+2P, N+3P,
!      etc., until it goes beyond the last scanline */
!   vI = arg->whichThread;
!   while (vI < arg->ctx->imgSize[1]) {
      if (nrrdCenterCell == arg->ctx->imgCentering) {
        v = uvScale*AIR_AFFINE(-0.5, vI, arg->ctx->imgSize[1]-0.5,
--- 199,220 ----
    }
  
!   while (1) {
!     /* the work assignment is simply the next scanline to be rendered:
!        the result of all this is setting vI */
!     if (arg->ctx->workMutex) {
!       airThreadMutexLock(arg->ctx->workMutex);
!     }
!     vI = arg->ctx->workIdx;
!     if (arg->ctx->workIdx < arg->ctx->imgSize[1]) {
!       arg->ctx->workIdx += 1;
!     }
!     if (arg->ctx->workMutex) {
!       airThreadMutexUnlock(arg->ctx->workMutex);
!     }
!     if (vI == arg->ctx->imgSize[1]) {
!       /* we're done! */
!       break;
!     }
! 
      if (nrrdCenterCell == arg->ctx->imgCentering) {
        v = uvScale*AIR_AFFINE(-0.5, vI, arg->ctx->imgSize[1]-0.5,
***************
*** 290,295 ****
        }
      }  /* end this scanline */
!     vI += arg->ctx->numThreads;
!   } /* end skipping through scanlines */
  
    if ( (ret = (arg->ctx->threadEnd)(thread,
--- 303,307 ----
        }
      }  /* end this scanline */
!   } /* end while(1) assignment of scanlines */
  
    if ( (ret = (arg->ctx->threadEnd)(thread,
***************
*** 353,356 ****
--- 365,374 ----
      thread[threadIdx] = airThreadNew();
    }
+   ctx->workIdx = 0;
+   if (1 < ctx->numThreads) {
+     ctx->workMutex = airThreadMutexNew();
+   } else {
+     ctx->workMutex = NULL;
+   }
  
    /* (done): call airThreadStart() once per thread, passing the
***************
*** 366,370 ****
       from this copy errArg->errCode into *errCodeP, and return
       errArg->whichErr */
! 
    if (1 < ctx->numThreads && !airThreadCapable) {
      fprintf(stderr, "%s: WARNING: not multi-threaded; will do %d "
--- 384,388 ----
       from this copy errArg->errCode into *errCodeP, and return
       errArg->whichErr */
!   
    if (1 < ctx->numThreads && !airThreadCapable) {
      fprintf(stderr, "%s: WARNING: not multi-threaded; will do %d "
***************
*** 397,400 ****
--- 415,422 ----
    }
  
+   if (1 < ctx->numThreads) {
+     ctx->workMutex = airThreadMutexNix(ctx->workMutex);
+   }
+ 
    if ( (ret = (ctx->renderEnd)(render, ctx->user)) ) {
      *errCodeP = ret;

Index: methodsHoover.c
===================================================================
RCS file: /cvsroot/teem/teem/src/hoover/methodsHoover.c,v
retrieving revision 1.10
retrieving revision 1.11
diff -C2 -d -r1.10 -r1.11
*** methodsHoover.c	10 May 2004 19:31:52 -0000	1.10
--- methodsHoover.c	11 May 2004 06:00:17 -0000	1.11
***************
*** 34,37 ****
--- 34,39 ----
      ctx->user = NULL;
      ctx->numThreads = 1;
+     ctx->workIdx = 0;
+     ctx->workMutex = NULL;
      ctx->renderBegin = hooverStubRenderBegin;
      ctx->threadBegin = hooverStubThreadBegin;
***************
*** 154,157 ****
--- 156,160 ----
    if (ctx) {
      limnCameraNix(ctx->cam);
+     /* workMutex is cleaned up at end of render */
      free(ctx);
    }