diff --git a/polybench-cuda/trmm/trmm.cu b/polybench-cuda/trmm/trmm.cu
index df4a7f72..84becf8b 100644
--- a/polybench-cuda/trmm/trmm.cu
+++ b/polybench-cuda/trmm/trmm.cu
@@ -22,6 +22,14 @@ void init_array(int n,int m,
   int i, j;
 
   *alpha = 32412;
+
+    // Initialize B to zero as only diagonals are set below
+  for (i = 0; i < m; i++) {
+    for (j = 0; j < n; j++) {
+      B[i*n + j] = 0.0;
+    }
+  }
+
   for (i = 0; i < n; i++)
     for (j = 0; j < m; j++) {
       A[i*m+j] = ((double) i*j) / m;
@@ -116,7 +124,7 @@ int main(int argc, char** argv)
   cudaMemcpy(dev_A, A, n*m*sizeof(double), cudaMemcpyHostToDevice);
   cudaMemcpy(dev_B, B, m*n*sizeof(double), cudaMemcpyHostToDevice);
   /* Run kernel. */
-  kernel(n,m, alpha, dev_A, dev_B);
+  kernel(n,m, alpha, dev_B, dev_A);
   cudaMemcpy(B, dev_B, m*n*sizeof(double), cudaMemcpyDeviceToHost);
 
   /* Prevent dead-code elimination. All live-out data must be printed
diff --git a/polybench-cuda/trmm/trmm_cpu.c b/polybench-cuda/trmm/trmm_cpu.c
index b94dea1e..ac76f3f6 100644
--- a/polybench-cuda/trmm/trmm_cpu.c
+++ b/polybench-cuda/trmm/trmm_cpu.c
@@ -153,7 +153,7 @@ int main(int argc, char ** argv) {
   B = malloc(m * n * 8);
   _ZL10init_arrayiiPdS_S_(n, m, (&alpha), ((double*)A), ((double*)B));
 ;
-  _ZL6kerneliidPdS_(n, m, alpha, ((double*)A), ((double*)B));
+  _ZL6kerneliidPdS_(n, m, alpha, ((double*)B), ((double*)A));
 ;
   if (dump_code == 1) {
 _ZL11print_arrayiiPd(m, n, ((double*)B));
@@ -167,10 +167,17 @@ free(((uint8_t*)((double*)B)));
 void _ZL10init_arrayiiPdS_S_(uint32_t n, uint32_t m, double* alpha, double* A, double* B) {
   int64_t i;
   int64_t j;
-  uint32_t j_2e_0;
+  uint32_t j_2e_1;
 
   *alpha = 32412;
 
+for(int64_t i = 0; i < m;   i = i + 1){
+
+for(int64_t j = 0; j < n;   j = j + 1){
+  B[(i * n + j)] = 0;
+}
+}
+
 for(int64_t i = 0; i < n;   i = i + 1){
 
 for(int64_t j = 0; j < m;   j = j + 1){