From 49ee6b02621ce3d7c1674b35dc7ccf61329d47ab Mon Sep 17 00:00:00 2001
From: Tanush Prathi <tanushp04@gmail.com>
Date: Tue, 8 Jul 2025 18:37:29 -0400
Subject: [PATCH] Fix kernels bug hopefully

---
 docs/documentation/readme.md       |  1 +
 src/simulation/m_data_output.fpp   | 14 +++++++-------
 src/simulation/m_time_steppers.fpp |  6 +++---
 3 files changed, 11 insertions(+), 10 deletions(-)

diff --git a/docs/documentation/readme.md b/docs/documentation/readme.md
index 267ab5ebf8..8bd2957332 100644
--- a/docs/documentation/readme.md
+++ b/docs/documentation/readme.md
@@ -9,6 +9,7 @@
 - [Running MFC](running.md)
 - [Flow Visualization](visualization.md)
 - [Performance](expectedPerformance.md)
+- [GPU Parallelization](gpuParallelization.md)
 - [MFC's Authors](authors.md)
 - [References](references.md)
 
diff --git a/src/simulation/m_data_output.fpp b/src/simulation/m_data_output.fpp
index 87a59c8866..dbf9f42552 100644
--- a/src/simulation/m_data_output.fpp
+++ b/src/simulation/m_data_output.fpp
@@ -316,14 +316,14 @@ contains
             Rc_min_loc = minval(Rc_sf)
         end if
 #else
-        !$acc kernels
-        icfl_max_loc = maxval(icfl_sf)
-        !$acc end kernels
+        #:call GPU_PARALLEL(copyout='[icfl_max_loc]', copyin='[icfl_sf]')
+            icfl_max_loc = maxval(icfl_sf)
+        #:endcall GPU_PARALLEL
         if (viscous) then
-            !$acc kernels
-            vcfl_max_loc = maxval(vcfl_sf)
-            Rc_min_loc = minval(Rc_sf)
-            !$acc end kernels
+            #:call GPU_PARALLEL(copyout='[vcfl_max_loc, Rc_min_loc]', copyin='[vcfl_sf,Rc_sf]')
+                vcfl_max_loc = maxval(vcfl_sf)
+                Rc_min_loc = minval(Rc_sf)
+            #:endcall GPU_PARALLEL
         end if
 #endif
 
diff --git a/src/simulation/m_time_steppers.fpp b/src/simulation/m_time_steppers.fpp
index 761ebdd792..53965cf464 100644
--- a/src/simulation/m_time_steppers.fpp
+++ b/src/simulation/m_time_steppers.fpp
@@ -993,9 +993,9 @@ contains
             end do
         end do
 
-        !$acc kernels
-        dt_local = minval(max_dt)
-        !$acc end kernels
+        #:call GPU_PARALLEL(copyout='[dt_local]', copyin='[max_dt]')
+            dt_local = minval(max_dt)
+        #:endcall GPU_PARALLEL
 
         if (num_procs == 1) then
             dt = dt_local