symm

polybench-c-4.2.1-beta/linear-algebra/blas/symm/symm_optimized.c

+/**
+ * This version is stamped on May 10, 2016
+ *
+ * Contact:
+ *   Louis-Noel Pouchet <pouchet.ohio-state.edu>
+ *   Tomofumi Yuki <tomofumi.yuki.fr>
+ *
+ * Web address: http://polybench.sourceforge.net
+ */
+/* symm.c: this file is part of PolyBench/C */
+
+#include <stdio.h>
+#include <unistd.h>
+#include <string.h>
+#include <math.h>
+
+/* Include polybench common header. */
+#include <polybench.h>
+
+/* Include benchmark-specific header. */
+#include "symm.h"
+
+
+/* Array initialization. */
+static
+void init_array(int m, int n,
+		DATA_TYPE *alpha,
+		DATA_TYPE *beta,
+		DATA_TYPE POLYBENCH_2D(C,M,N,m,n),
+		DATA_TYPE POLYBENCH_2D(A,M,M,m,m),
+		DATA_TYPE POLYBENCH_2D(B,M,N,m,n))
+{
+  int i, j;
+
+  *alpha = 1.5;
+  *beta = 1.2;
+  for (i = 0; i < m; i++)
+    for (j = 0; j < n; j++) {
+      C[i][j] = (DATA_TYPE) ((i+j) % 100) / m;
+      B[i][j] = (DATA_TYPE) ((n+i-j) % 100) / m;
+    }
+  for (i = 0; i < m; i++) {
+    for (j = 0; j <=i; j++)
+      A[i][j] = (DATA_TYPE) ((i+j) % 100) / m;
+    for (j = i+1; j < m; j++)
+      A[i][j] = -999; //regions of arrays that should not be used
+  }
+}
+
+
+/* DCE code. Must scan the entire live-out data.
+   Can be used also to check the correctness of the output. */
+static
+void print_array(int m, int n,
+		 DATA_TYPE POLYBENCH_2D(C,M,N,m,n))
+{
+  int i, j;
+
+  POLYBENCH_DUMP_START;
+  POLYBENCH_DUMP_BEGIN("C");
+  for (i = 0; i < m; i++)
+    for (j = 0; j < n; j++) {
+	if ((i * m + j) % 20 == 0) fprintf (POLYBENCH_DUMP_TARGET, "\n");
+	fprintf (POLYBENCH_DUMP_TARGET, DATA_PRINTF_MODIFIER, C[i][j]);
+    }
+  POLYBENCH_DUMP_END("C");
+  POLYBENCH_DUMP_FINISH;
+}
+
+
+/* Main computational kernel. The whole function will be timed,
+   including the call and return. */
+static
+void kernel_symm(int m, int n,
+		 DATA_TYPE alpha,
+		 DATA_TYPE beta,
+		 DATA_TYPE POLYBENCH_2D(C,M,N,m,n),
+		 DATA_TYPE POLYBENCH_2D(A,M,M,m,m),
+		 DATA_TYPE POLYBENCH_2D(B,M,N,m,n))
+{
+  int i, j, k;
+  DATA_TYPE temp2;
+
+//BLAS PARAMS
+//SIDE = 'L'
+//UPLO = 'L'
+// =>  Form  C := alpha*A*B + beta*C
+// A is MxM
+// B is MxN
+// C is MxN
+//note that due to Fortran array layout, the code below more closely resembles upper triangular case in BLAS
+#pragma scop
+   for (i = 0; i < _PB_M; i++){
+      
+      #pragma omp parallel for
+      for (j = 0; j < _PB_N; j++ )
+        for (k = 0; k < i; k++) 
+           C[k][j] += alpha*B[i][j] * A[i][k];
+
+      for (j = 0; j < _PB_N; j++ ){
+        temp2 = 0;
+        for (k = 0; k < i; k++) {
+           temp2 += B[k][j] * A[i][k];
+        }
+        C[i][j] = beta * C[i][j] + alpha*B[i][j] * A[i][i] + alpha * temp2;
+     }
+     
+   }
+#pragma endscop
+
+}
+
+
+int main(int argc, char** argv)
+{
+  /* Retrieve problem size. */
+  int m = M;
+  int n = N;
+
+  /* Variable declaration/allocation. */
+  DATA_TYPE alpha;
+  DATA_TYPE beta;
+  POLYBENCH_2D_ARRAY_DECL(C,DATA_TYPE,M,N,m,n);
+  POLYBENCH_2D_ARRAY_DECL(A,DATA_TYPE,M,M,m,m);
+  POLYBENCH_2D_ARRAY_DECL(B,DATA_TYPE,M,N,m,n);
+
+  /* Initialize array(s). */
+  init_array (m, n, &alpha, &beta,
+	      POLYBENCH_ARRAY(C),
+	      POLYBENCH_ARRAY(A),
+	      POLYBENCH_ARRAY(B));
+
+  /* Start timer. */
+  polybench_start_instruments;
+
+  /* Run kernel. */
+  kernel_symm (m, n,
+	       alpha, beta,
+	       POLYBENCH_ARRAY(C),
+	       POLYBENCH_ARRAY(A),
+	       POLYBENCH_ARRAY(B));
+
+  /* Stop and print timer. */
+  polybench_stop_instruments;
+  polybench_print_instruments;
+
+  /* Prevent dead-code elimination. All live-out data must be printed
+     by the function call in argument. */
+  polybench_prevent_dce(print_array(m, n, POLYBENCH_ARRAY(C)));
+
+  /* Be clean. */
+  POLYBENCH_FREE_ARRAY(C);
+  POLYBENCH_FREE_ARRAY(A);
+  POLYBENCH_FREE_ARRAY(B);
+
+  return 0;
+}