readd memory fences

olepoeschl · Feb 19, 2024 · d2864dd · d2864dd
1 parent 5bce312
commit d2864dd
Show file tree

Hide file tree

Showing 2 changed files with 16 additions and 13 deletions.
diff --git a/src/main/java/de/nqueensfaf/Demo.java b/src/main/java/de/nqueensfaf/Demo.java
@@ -12,17 +12,17 @@ public static void main(String[] args) {
 	gpu();
     }
 
-    static void cpu() {
-	new CpuSolver()
-        	.setPresetQueens(5)
-        	.setThreadCount(1)
-        	.setUpdateInterval(800)
-        	.onInit(self -> System.out.println("Starting Solver for board size " + self.getN() + "..."))
-        	.onUpdate((self, progress, solutions, duration) -> System.out.println("progress: " + progress + " solutions: " + solutions + " duration: " + duration))
-        	.onFinish(self -> System.out.println("Found " + self.getSolutions() + " solutions in " + self.getDuration() + " ms"))
-        	.setN(16)
-        	.solve();
-    }
+//    static void cpu() {
+//	new CpuSolver()
+//        	.setPresetQueens(5)
+//        	.setThreadCount(1)
+//        	.setUpdateInterval(800)
+//        	.onInit(self -> System.out.println("Starting Solver for board size " + self.getN() + "..."))
+//        	.onUpdate((self, progress, solutions, duration) -> System.out.println("progress: " + progress + " solutions: " + solutions + " duration: " + duration))
+//        	.onFinish(self -> System.out.println("Found " + self.getSolutions() + " solutions in " + self.getDuration() + " ms"))
+//        	.setN(16)
+//        	.solve();
+//    }
 
     static void gpu() {
 	GpuSolver g = new GpuSolver();
@@ -31,14 +31,14 @@ static void gpu() {
 	    if(gpu.vendor().toLowerCase().contains("nvidia")) {
 		g.gpuSelection().add(gpu.id(), 5, 64);
 	    } else {
-//		g.gpuSelection().add(gpu.id(), 50, 64);
+		g.gpuSelection().add(gpu.id(), 50, 24);
 	    }
 	}
 	g.setUpdateInterval(400);
 	g.onInit(self -> System.out.println("Starting Solver for board size " + self.getN() + "..."))
         	.onUpdate((self, progress, solutions, duration) -> System.out.println("progress: " + progress + " solutions: " + solutions + " duration: " + duration))
         	.onFinish(self -> System.out.println("Found " + self.getSolutions() + " solutions in " + self.getDuration() + " ms"))
-		.setN(18)
+		.setN(20)
 		.solve();
 
 //	new GPUSolver()

diff --git a/src/main/resources/kernels.c b/src/main/resources/kernels.c
@@ -34,6 +34,7 @@ kernel void nqfaf_nvidia(global struct constellation *constellation_arr, global
     // in row k only L is free and in row l only 1 is free 
     local uint jkl_queens[N];
     jkl_queens[l_id % N] = jkl_queens_arr[get_group_id(0) * N + l_id % N];
+    barrier(CLK_LOCAL_MEM_FENCE);
 
     uint ldiag = L >> ((c.start_ijkl >> 5) & 31); // ld from queen l with respect to the first row 
     uint rdiag = 1 << (c.start_ijkl & 31); // ld from queen k with respect to the first row 
@@ -205,6 +206,7 @@ kernel void nqfaf_amd(constant struct constellation *constellation_arr, global u
     // in row k only L is free and in row l only 1 is free 
     local uint jkl_queens[N];
     jkl_queens[l_id % N] = jkl_queens_arr[get_group_id(0) * N + l_id % N];
+    barrier(CLK_LOCAL_MEM_FENCE);
 
     uint ldiag = L >> ((c.start_ijkl >> 5) & 31); // ld from queen l with respect to the first row 
     uint rdiag = 1 << (c.start_ijkl & 31); // ld from queen k with respect to the first row 
@@ -376,6 +378,7 @@ kernel void nqfaf_intel(global struct constellation *constellation_arr, global u
     // in row k only L is free and in row l only 1 is free 
     local uint jkl_queens[N];
     jkl_queens[l_id % N] = jkl_queens_arr[get_group_id(0) * N + l_id % N];
+    barrier(CLK_LOCAL_MEM_FENCE);
 
     uint ldiag = L >> ((c.start_ijkl >> 5) & 31); // ld from queen l with respect to the first row 
     uint rdiag = 1 << (c.start_ijkl & 31); // ld from queen k with respect to the first row