[tests] Add compute-bound variant of vecadd

This loops 1000 times over `sum += A[i] + B[i]`, making every memory op
hit at L1 cache.

tests/opencl/vecadd/kernel.alll1hit.loop1000.cl

@@ -0,0 +1,12 @@
+__kernel void vecadd (__global const float *A,
+	                    __global const float *B,
+	                    __global float *C)
+{
+  int gid = get_global_id(0);
+  float sum = 0.;
+  for (int i = 0; i < 1000; i++) {
+	  int addr = gid + (i % 2);
+	  sum += A[addr] + B[addr];
+  }
+  C[gid] = sum;
+}