use example mnemonics, fix in Examples async_target and task_dep

Examples_async_target.tex

@@ -43,6 +43,12 @@ vectors point to the correct local storage, of the space that was not freed in t
 directive.  At the end of the second \code{target} region, the data in array \plc{p} is copied back
 to the host since \plc{p} is not an allocatable array.
 
+A \code{depend} clause is used in the \code{task} directive to provide a wait at the beginning of the second 
+\code{target} region, to insure that there is no race condition with \plc{v1} and \plc{v2} in the two tasks.
+It would be noncompliant to use \plc{v1} and/or \plc{v2} in lieu of \plc{N} in the \code{depend} clauses, 
+because the use of non-allocated allocatable arrays as list items in the first \code{depend} clause would 
+lead to unspecified behavior. 
+
 \fexample{async_target}{2f}
 
 

openmp.sty

@@ -439,9 +439,9 @@
    }{
       \def\cname{#1.#2}
 % Use following line for old numbering
-      \def\ename{\thechapter.#2}
+%      \def\ename{\thechapter.#2}
 % Use following for mneumonics
-%      \def\ename{\escstr{#1}.#2}
+      \def\ename{\escstr{#1}.#2}
    }
    \noindent
    \textit{Example \ename}

sources/Example_async_target.2c.c

@@ -39,5 +39,6 @@ void vec_mult(float *p, int N, int dev)
        free(v1);
        free(v2);
    }
+   #pragma omp taskwait
    output(p, N);
 }

sources/Example_async_target.2f.f

@@ -10,7 +10,7 @@
    integer ::  i, idev
    !$omp declare target (init)
 
-   !$omp task shared(v1,v2) depend(out: v1,v2)
+   !$omp task shared(v1,v2) depend(out: N)
       !$omp target device(idev) map(v1,v2)
          if( omp_is_initial_device() ) &
             stop "not executing on target device"
@@ -21,7 +21,7 @@
 
    call foo()  ! execute other work asychronously
 
-   !$omp task shared(v1,v2,p) depend(in: v1,v2)
+   !$omp task shared(v1,v2,p) depend(in: N)
       !$omp target device(idev) map(to: v1,v2) map(from: p)
          if( omp_is_initial_device() ) &
             stop "not executing on target device"
@@ -34,6 +34,7 @@
       !$omp end target
    !$omp end task
 
+   !$omp taskwait
    call output(p, N)
 
 end subroutine

sources/Example_task_dep.5c.c

@@ -13,7 +13,10 @@ C[N][N] )
    for (i = 0; i < N; i+=BS) {
       for (j = 0; j < N; j+=BS) {
          for (k = 0; k < N; k+=BS) {
-#pragma omp task depend ( in: A[i:BS][k:BS], B[k:BS][j:BS] ) \
+// Note 1: i, j, k, A, B, C are firstprivate by default
+// Note 2: A, B and C are just pointers
+#pragma omp task private(ii, jj, kk) \
+            depend ( in: A[i:BS][k:BS], B[k:BS][j:BS] ) \
             depend ( inout: C[i:BS][j:BS] )
             for (ii = i; ii < i+BS; ii++ )
                for (jj = j; jj < j+BS; jj++ )

sources/Example_task_dep.5f.f

@@ -3,19 +3,22 @@
 ! @@compilable:	yes
 ! @@linkable:	no
 ! @@expect:	success
+! Assume BS divides N perfectly
 subroutine matmul_depend (N, BS, A, B, C)
+   implicit none
    integer :: N, BS, BM
    real, dimension(N, N) :: A, B, C
    integer :: i, j, k, ii, jj, kk
-   BM = BS -1
+   BM = BS - 1
    do i = 1, N, BS
       do j = 1, N, BS
          do k = 1, N, BS
-!$omp task depend ( in: A(i:i+BM, k:k+BM), B(k:k+BM, j:j+BM) ) &
+!$omp task shared(A,B,C) private(ii,jj,kk) & ! I,J,K are firstprivate by default
+!$omp depend ( in: A(i:i+BM, k:k+BM), B(k:k+BM, j:j+BM) ) &
 !$omp depend ( inout: C(i:i+BM, j:j+BM) )
-            do ii = i, i+BS
+            do ii = i, i+BM
-               do jj = j, j+BS
+               do jj = j, j+BM
-                  do kk = k, k+BS
+                  do kk = k, k+BM
                      C(jj,ii) = C(jj,ii) + A(kk,ii) * B(jj,kk)
                   end do
                end do