From 16f92a4dd2776dc59215e4b52e8e63c328086fa6 Mon Sep 17 00:00:00 2001
From: =?UTF-8?q?Ole=20P=C3=B6schl?=
 <62295828+olepoeschl@users.noreply.github.com>
Date: Wed, 21 Jun 2023 13:02:58 +0200
Subject: [PATCH] Added barriers to nvidia and amd kernels: local inside the
 main loop and global at the end after writing the result

refs #12
---
 src/main/resources/kernels.c | 4 ++++
 1 file changed, 4 insertions(+)

diff --git a/src/main/resources/kernels.c b/src/main/resources/kernels.c
index c283fa56..32dcbb4a 100644
--- a/src/main/resources/kernels.c
+++ b/src/main/resources/kernels.c
@@ -95,6 +95,7 @@ kernel void nqfaf_nvidia(global int *ld_arr, global int *rd_arr, global int *col
 			ld_mem >>= 1;
 			rd_mem <<= 1;						
 		}
+		barrier(CLK_LOCAL_MEM_FENCE);
 		free = ~(jkl_queens[row] | ld | rd | col);		// calculate the occupancy of the next row
 		free &= ~(queen + direction-1);					// occupy all bits right from the last queen in order to not place the same queen again 
 		col ^= queen;									// free up the column AFTER calculating free in order to not place the same queen again		
@@ -103,6 +104,7 @@ kernel void nqfaf_nvidia(global int *ld_arr, global int *rd_arr, global int *col
 			solutions++;
 	}
 	result[g_id] = solutions;						// number of solutions of the work item 
+	barrier(CLK_GLOBAL_MEM_FENCE);
 }
 
 // AMD kernel
@@ -194,6 +196,7 @@ kernel void nqfaf_amd(global int *ld_arr, global int *rd_arr, global int *col_ar
 			ld_mem >>= 1;
 			rd_mem <<= 1;						
 		}
+		barrier(CLK_LOCAL_MEM_FENCE);
 		free = ~(jkl_queens[row] | ld | rd | col); 		// calculate the occupancy of the next row
 		free &= ~(queen + direction-1);					// occupy all bits right from the last queen in order to not place the same queen again 
 		col ^= queen;									// free up the column AFTER calculating free in order to not place the same queen again		
@@ -201,6 +204,7 @@ kernel void nqfaf_amd(global int *ld_arr, global int *rd_arr, global int *col_ar
 		solutions += (row == N-1);						// increase the solutions, if we are in the last row 
 	}
 	result[g_id] = solutions;						// number of solutions of the work item 
+	barrier(CLK_GLOBAL_MEM_FENCE);
 }
 
 // Intel kernel