Wrapper init_us_2 now receives vkb (on CPU)

PW/src/c_bands.f90

@@ -111,13 +111,19 @@ SUBROUTINE c_bands( iter )
      !
      ! ... More stuff needed by the hamiltonian: nonlocal projectors
      !
-     IF ( use_gpu ) THEN
-        IF ( nkb > 0 ) CALL using_vkb_d(2)
-        IF ( nkb > 0 ) CALL init_us_2_gpu( ngk(ik), igk_k_d(1,ik), xk(1,ik), vkb_d )
-     ELSE
-        IF ( nkb > 0 ) CALL using_vkb(2)
-        IF ( nkb > 0 ) CALL init_us_2( ngk(ik), igk_k(1,ik), xk(1,ik), vkb )
+!civn 
+!    IF ( use_gpu ) THEN
+!       IF ( nkb > 0 ) CALL using_vkb_d(2)
+!       IF ( nkb > 0 ) CALL init_us_2_gpu( ngk(ik), igk_k_d(1,ik), xk(1,ik), vkb_d )
+!    ELSE
+!       IF ( nkb > 0 ) CALL using_vkb(2)
+!       IF ( nkb > 0 ) CALL init_us_2( ngk(ik), igk_k(1,ik), xk(1,ik), vkb )
+!    END IF
+     IF ( nkb > 0 ) THEN
+       CALL using_vkb(2)
+       CALL init_us_2( ngk(ik), igk_k(1,ik), xk(1,ik), vkb )
      END IF
+!
      !
      ! ... read in wavefunctions from the previous iteration
      !

PW/src/init_us_2.f90

@@ -33,9 +33,21 @@ SUBROUTINE init_us_2( npw_, igk_, q_, vkb_ )
   !
   CALL start_clock( 'init_us_2' )
   !
+!civn 
+#if defined(__CUDA)
+!$acc data copyin(igk_(npw_), eigts1(:,:), eigts2(:,:), eigts3(:,:), mill(:,:), g(:,:)) copy(vkb_(npwx,nkb))
+!$acc host_data use_device(eigts1, eigts2, eigts3, mill, g, igk_, vkb_)
+  CALL init_us_2_base_gpu(npw_, npwx, igk_, q_, nat, tau, ityp, tpiba, omega,&
+    dfftp%nr1, dfftp%nr2, dfftp%nr3, eigts1, eigts2, eigts3, mill, g,&
+    vkb_ )
+!$acc end host_data
+!$acc end data
+#else
   CALL init_us_2_base(npw_, npwx, igk_, q_, nat, tau, ityp, tpiba, omega, &
           dfftp%nr1, dfftp%nr2, dfftp%nr3, eigts1, eigts2, eigts3, mill, g,&
           vkb_ )
+#endif
+!
   !
   CALL stop_clock( 'init_us_2' )
   !

upflib/init_us_2_base.f90

@@ -67,6 +67,9 @@ SUBROUTINE init_us_2_base( npw_, npwx, igk_, q_, nat, tau, ityp, &
   INTEGER, PARAMETER :: blocksize = 256
   INTEGER     :: iblock, numblock, realblocksize
   !
+!civn 
+write(*,*) "using init_us_2_base"  
+!
   IF (lmaxkb < 0) RETURN
   !
   ! setting cache blocking size

upflib/init_us_2_base_gpu.f90

@@ -82,6 +82,9 @@ SUBROUTINE init_us_2_base_gpu( npw_, npwx, igk__d, q_, nat, tau, ityp, &
 #endif
   !
   !
+!civn 
+write(*,*) "using init_us_2_base_gpu"  
+!
   if (lmaxkb<0) return
   
   ! JR Eventually replace with smarter allocation/deallocation of GPU temp arrays