reduce size of collisions[] to 1 (avg. +7% speedup observed)

mbevand · Nov 10, 2016 · ba0dd19 · ba0dd19
1 parent 5228f1d
commit ba0dd19
Show file tree

Hide file tree

Showing 2 changed files with 3 additions and 4 deletions.
diff --git a/README.md b/README.md
@@ -85,9 +85,8 @@ and statistics in progressively more and more details.
 
 # Performance
 
-* 47.5 sol/s with one R9 Nano
-* 45.0 sol/s with one R9 290X
-* 41.0 sol/s with one RX 480 8GB
+* 51.0 sol/s with one R9 Nano
+* 44.0 sol/s with one RX 480 8GB
 * 30.5 sol/s with one GTX Titan X (Maxwell)
 * 30.5 sol/s with one GTX Titan (Kepler)
 

diff --git a/input.cl b/input.cl
@@ -703,7 +703,7 @@ void kernel_sols(__global char *ht0, __global char *ht1, __global sols_t *sols)
     uint		ref_i, ref_j;
     // it's ok for the collisions array to be so small, as if it fills up
     // the potential solutions are likely invalid (many duplicate inputs)
-    ulong		collisions[5];
+    ulong		collisions[1];
     uint		coll;
 #if NR_ROWS_LOG >= 16 && NR_ROWS_LOG <= 20
     // in the final hash table, we are looking for a match on both the bits