Index: test-suite/trunk/SingleSource/Benchmarks/Polybench/datamining/correlation/correlation.c
===================================================================
--- test-suite/trunk/SingleSource/Benchmarks/Polybench/datamining/correlation/correlation.c
+++ test-suite/trunk/SingleSource/Benchmarks/Polybench/datamining/correlation/correlation.c
@@ -43,11 +43,12 @@
 
 {
   int i, j;
-  char *printmat = malloc(m*8);
+  char *printmat = malloc(m*16 + 1); printmat[m*16] = 0;
 
   for (i = 0; i < m; i++) {
-    for (j = 0; j < m; j++)
-      print_element(symmat[i][j], j*8, printmat);
+    for (j = 0; j < m; j++) {
+      print_element(symmat[i][j], j*16, printmat);
+    }
     fputs(printmat, stderr);
   }
   free(printmat);
Index: test-suite/trunk/SingleSource/Benchmarks/Polybench/datamining/correlation/correlation.reference_output
===================================================================
--- test-suite/trunk/SingleSource/Benchmarks/Polybench/datamining/correlation/correlation.reference_output
+++ test-suite/trunk/SingleSource/Benchmarks/Polybench/datamining/correlation/correlation.reference_output
@@ -1 +1 @@
-ab44d09216ee09fe72e448e7c9a0a13b
+1ba44a2636ee1214d9358c8a1a9890ef
Index: test-suite/trunk/SingleSource/Benchmarks/Polybench/datamining/correlation/correlation.reference_output.small
===================================================================
--- test-suite/trunk/SingleSource/Benchmarks/Polybench/datamining/correlation/correlation.reference_output.small
+++ test-suite/trunk/SingleSource/Benchmarks/Polybench/datamining/correlation/correlation.reference_output.small
@@ -1 +1 @@
-4dcdd4b95a19a6cfbd32f4d5f24e17be
+e8a2fb6cb92e6edbf300ecdacd647316
Index: test-suite/trunk/SingleSource/Benchmarks/Polybench/datamining/covariance/covariance.c
===================================================================
--- test-suite/trunk/SingleSource/Benchmarks/Polybench/datamining/covariance/covariance.c
+++ test-suite/trunk/SingleSource/Benchmarks/Polybench/datamining/covariance/covariance.c
@@ -42,11 +42,11 @@
 
 {
   int i, j;
-  char *printmat = malloc(m*8);
+  char *printmat = malloc(m*16 + 1); printmat[m*16] = 0;
 
   for (i = 0; i < m; i++) {
     for (j = 0; j < m; j++)
-      print_element(symmat[i][j], j*8, printmat);
+      print_element(symmat[i][j], j*16, printmat);
     fputs(printmat, stderr);
   }
   free(printmat);
Index: test-suite/trunk/SingleSource/Benchmarks/Polybench/datamining/covariance/covariance.reference_output
===================================================================
--- test-suite/trunk/SingleSource/Benchmarks/Polybench/datamining/covariance/covariance.reference_output
+++ test-suite/trunk/SingleSource/Benchmarks/Polybench/datamining/covariance/covariance.reference_output
@@ -1 +1 @@
-7a284ecb4d8b0f9e4132181b8c8754cc
+82b48e22752b25e42e5ba7a456c07467
Index: test-suite/trunk/SingleSource/Benchmarks/Polybench/datamining/covariance/covariance.reference_output.small
===================================================================
--- test-suite/trunk/SingleSource/Benchmarks/Polybench/datamining/covariance/covariance.reference_output.small
+++ test-suite/trunk/SingleSource/Benchmarks/Polybench/datamining/covariance/covariance.reference_output.small
@@ -1 +1 @@
-cd7b8424a3c3635d1ea90f42e415faa7
+63a8ee32bee3b68a5edc4fd680ed438d
Index: test-suite/trunk/SingleSource/Benchmarks/Polybench/linear-algebra/kernels/2mm/2mm.c
===================================================================
--- test-suite/trunk/SingleSource/Benchmarks/Polybench/linear-algebra/kernels/2mm/2mm.c
+++ test-suite/trunk/SingleSource/Benchmarks/Polybench/linear-algebra/kernels/2mm/2mm.c
@@ -54,11 +54,11 @@
 		 DATA_TYPE POLYBENCH_2D(D,NI,NL,ni,nl))
 {
   int i, j;
-  char *printmat = malloc(nl*8);
+  char *printmat = malloc(nl*16 + 1); printmat[nl*16] = 0;
 
   for (i = 0; i < ni; i++) {
     for (j = 0; j < nl; j++)
-      print_element(D[i][j], j*8, printmat);
+      print_element(D[i][j], j*16, printmat);
     fputs(printmat, stderr);
   }
   free(printmat);
Index: test-suite/trunk/SingleSource/Benchmarks/Polybench/linear-algebra/kernels/2mm/2mm.reference_output
===================================================================
--- test-suite/trunk/SingleSource/Benchmarks/Polybench/linear-algebra/kernels/2mm/2mm.reference_output
+++ test-suite/trunk/SingleSource/Benchmarks/Polybench/linear-algebra/kernels/2mm/2mm.reference_output
@@ -1 +1 @@
-f9f294040187318bc2cc477e7eecab01
+422aa03845e5d0cd97cb018c5b4d61e5
Index: test-suite/trunk/SingleSource/Benchmarks/Polybench/linear-algebra/kernels/2mm/2mm.reference_output.small
===================================================================
--- test-suite/trunk/SingleSource/Benchmarks/Polybench/linear-algebra/kernels/2mm/2mm.reference_output.small
+++ test-suite/trunk/SingleSource/Benchmarks/Polybench/linear-algebra/kernels/2mm/2mm.reference_output.small
@@ -1 +1 @@
-68efcbd188b404f4fb09803b61c02f60
+eac6d4f27d66fcd1d9e32bc3155f93b6
Index: test-suite/trunk/SingleSource/Benchmarks/Polybench/linear-algebra/kernels/3mm/3mm.c
===================================================================
--- test-suite/trunk/SingleSource/Benchmarks/Polybench/linear-algebra/kernels/3mm/3mm.c
+++ test-suite/trunk/SingleSource/Benchmarks/Polybench/linear-algebra/kernels/3mm/3mm.c
@@ -50,11 +50,11 @@
 		 DATA_TYPE POLYBENCH_2D(G,NI,NL,ni,nl))
 {
   int i, j;
-  char *printmat = malloc(nl*8);
+  char *printmat = malloc(nl*16 + 1); printmat[nl*16] = 0;
 
   for (i = 0; i < ni; i++) {
     for (j = 0; j < nl; j++)
-      print_element(G[i][j], j*8, printmat);
+      print_element(G[i][j], j*16, printmat);
     fputs(printmat, stderr);
   }
   free(printmat);
Index: test-suite/trunk/SingleSource/Benchmarks/Polybench/linear-algebra/kernels/3mm/3mm.reference_output
===================================================================
--- test-suite/trunk/SingleSource/Benchmarks/Polybench/linear-algebra/kernels/3mm/3mm.reference_output
+++ test-suite/trunk/SingleSource/Benchmarks/Polybench/linear-algebra/kernels/3mm/3mm.reference_output
@@ -1 +1 @@
-f52121e0bccc072008e52a06798e05db
+bddaec967faf18122e44b0073e6f8532
Index: test-suite/trunk/SingleSource/Benchmarks/Polybench/linear-algebra/kernels/3mm/3mm.reference_output.small
===================================================================
--- test-suite/trunk/SingleSource/Benchmarks/Polybench/linear-algebra/kernels/3mm/3mm.reference_output.small
+++ test-suite/trunk/SingleSource/Benchmarks/Polybench/linear-algebra/kernels/3mm/3mm.reference_output.small
@@ -1 +1 @@
-7779d064d80186d46a0d91b8f6116c62
+5649bd55b3ddd097acb8ebb6cb14b3b0
Index: test-suite/trunk/SingleSource/Benchmarks/Polybench/linear-algebra/kernels/atax/atax.c
===================================================================
--- test-suite/trunk/SingleSource/Benchmarks/Polybench/linear-algebra/kernels/atax/atax.c
+++ test-suite/trunk/SingleSource/Benchmarks/Polybench/linear-algebra/kernels/atax/atax.c
@@ -42,10 +42,10 @@
 
 {
   int i;
-  char *printmat = malloc(nx*8);
+  char *printmat = malloc(nx*16 + 1); printmat[nx*16] = 0;
 
   for (i = 0; i < nx; i++)
-    print_element(y[i], i*8, printmat);
+    print_element(y[i], i*16, printmat);
   fputs(printmat, stderr);
   free(printmat);
 }
Index: test-suite/trunk/SingleSource/Benchmarks/Polybench/linear-algebra/kernels/atax/atax.reference_output
===================================================================
--- test-suite/trunk/SingleSource/Benchmarks/Polybench/linear-algebra/kernels/atax/atax.reference_output
+++ test-suite/trunk/SingleSource/Benchmarks/Polybench/linear-algebra/kernels/atax/atax.reference_output
@@ -1 +1 @@
-88a04eb3d01f6346a6ee4b678e311fe5
+55df10ccc646517c6e78304b428a5b53
Index: test-suite/trunk/SingleSource/Benchmarks/Polybench/linear-algebra/kernels/bicg/bicg.c
===================================================================
--- test-suite/trunk/SingleSource/Benchmarks/Polybench/linear-algebra/kernels/bicg/bicg.c
+++ test-suite/trunk/SingleSource/Benchmarks/Polybench/linear-algebra/kernels/bicg/bicg.c
@@ -47,14 +47,14 @@
 {
   int i;
   int n = nx > ny ? nx : ny;
-  char *printmat = malloc(n*8);
+  char *printmat = malloc(n*16 + 1); printmat[n*16] = 0;
 
   for (i = 0; i < ny; i++)
-    print_element(s[i], i*8, printmat);
+    print_element(s[i], i*16, printmat);
   *(printmat+i) = 0;
   fputs(printmat, stderr);
   for (i = 0; i < nx; i++)
-    print_element(q[i], i*8, printmat);
+    print_element(q[i], i*16, printmat);
   *(printmat+i) = 0;
   fputs(printmat, stderr);
   free(printmat);
Index: test-suite/trunk/SingleSource/Benchmarks/Polybench/linear-algebra/kernels/bicg/bicg.reference_output
===================================================================
--- test-suite/trunk/SingleSource/Benchmarks/Polybench/linear-algebra/kernels/bicg/bicg.reference_output
+++ test-suite/trunk/SingleSource/Benchmarks/Polybench/linear-algebra/kernels/bicg/bicg.reference_output
@@ -1 +1 @@
-22fab907b016327a2808510b55959131
+74944d07ab36d6b25801a785eae90cac
Index: test-suite/trunk/SingleSource/Benchmarks/Polybench/linear-algebra/kernels/cholesky/cholesky.c
===================================================================
--- test-suite/trunk/SingleSource/Benchmarks/Polybench/linear-algebra/kernels/cholesky/cholesky.c
+++ test-suite/trunk/SingleSource/Benchmarks/Polybench/linear-algebra/kernels/cholesky/cholesky.c
@@ -54,11 +54,11 @@
 
 {
   int i, j;
-  char *printmat = malloc(n*8);
+  char *printmat = malloc(n*16 + 1); printmat[n*16] = 0;
 
   for (i = 0; i < n; i++) {
     for (j = 0; j < n; j++)
-      print_element(A[i][j], j*8, printmat);
+      print_element(A[i][j], j*16, printmat);
     fputs(printmat, stderr);
   }
   free(printmat);
Index: test-suite/trunk/SingleSource/Benchmarks/Polybench/linear-algebra/kernels/cholesky/cholesky.reference_output
===================================================================
--- test-suite/trunk/SingleSource/Benchmarks/Polybench/linear-algebra/kernels/cholesky/cholesky.reference_output
+++ test-suite/trunk/SingleSource/Benchmarks/Polybench/linear-algebra/kernels/cholesky/cholesky.reference_output
@@ -1 +1 @@
-5da79cc798c356034dfdf5ac98b3ecaa
+2adac2da4878b10c3a2c0881ecb534ac
Index: test-suite/trunk/SingleSource/Benchmarks/Polybench/linear-algebra/kernels/cholesky/cholesky.reference_output.small
===================================================================
--- test-suite/trunk/SingleSource/Benchmarks/Polybench/linear-algebra/kernels/cholesky/cholesky.reference_output.small
+++ test-suite/trunk/SingleSource/Benchmarks/Polybench/linear-algebra/kernels/cholesky/cholesky.reference_output.small
@@ -1 +1 @@
-e9aaf9cf2963bd1cc5183174c0dfadc1
+637d089a6787bb1a31273d62104fbee1
Index: test-suite/trunk/SingleSource/Benchmarks/Polybench/linear-algebra/kernels/doitgen/doitgen.c
===================================================================
--- test-suite/trunk/SingleSource/Benchmarks/Polybench/linear-algebra/kernels/doitgen/doitgen.c
+++ test-suite/trunk/SingleSource/Benchmarks/Polybench/linear-algebra/kernels/doitgen/doitgen.c
@@ -43,12 +43,12 @@
 		 DATA_TYPE POLYBENCH_3D(A,NR,NQ,NP,nr,nq,np))
 {
   int i, j, k;
-  char *printmat = malloc(np*8);
+  char *printmat = malloc(np*16 + 1); printmat[np*16] = 0;
 
   for (i = 0; i < nr; i++) {
     for (j = 0; j < nq; j++)
       for (k = 0; k < np; k++)
-      print_element(A[i][j][k], k*8, printmat);
+      print_element(A[i][j][k], k*16, printmat);
     fputs(printmat, stderr);
   }
   free(printmat);
Index: test-suite/trunk/SingleSource/Benchmarks/Polybench/linear-algebra/kernels/doitgen/doitgen.reference_output
===================================================================
--- test-suite/trunk/SingleSource/Benchmarks/Polybench/linear-algebra/kernels/doitgen/doitgen.reference_output
+++ test-suite/trunk/SingleSource/Benchmarks/Polybench/linear-algebra/kernels/doitgen/doitgen.reference_output
@@ -1 +1 @@
-73c37cc505bf1b3706695ed80db85dbf
+fbacccfd2d72d91c5287be840c949cec
Index: test-suite/trunk/SingleSource/Benchmarks/Polybench/linear-algebra/kernels/doitgen/doitgen.reference_output.small
===================================================================
--- test-suite/trunk/SingleSource/Benchmarks/Polybench/linear-algebra/kernels/doitgen/doitgen.reference_output.small
+++ test-suite/trunk/SingleSource/Benchmarks/Polybench/linear-algebra/kernels/doitgen/doitgen.reference_output.small
@@ -1 +1 @@
-3c8a6a1e9fe6738b6d2d79335360194d
+8a5318ea2236f721194816e9aea7fd56
Index: test-suite/trunk/SingleSource/Benchmarks/Polybench/linear-algebra/kernels/gemm/gemm.c
===================================================================
--- test-suite/trunk/SingleSource/Benchmarks/Polybench/linear-algebra/kernels/gemm/gemm.c
+++ test-suite/trunk/SingleSource/Benchmarks/Polybench/linear-algebra/kernels/gemm/gemm.c
@@ -50,11 +50,11 @@
 		 DATA_TYPE POLYBENCH_2D(C,NI,NJ,ni,nj))
 {
   int i, j;
-  char *printmat = malloc(nj*8);
+  char *printmat = malloc(nj*16 + 1); printmat[nj*16] = 0;
 
   for (i = 0; i < ni; i++) {
     for (j = 0; j < nj; j++)
-      print_element(C[i][j], j*8, printmat);
+      print_element(C[i][j], j*16, printmat);
     fputs(printmat, stderr);
   }
   free(printmat);
Index: test-suite/trunk/SingleSource/Benchmarks/Polybench/linear-algebra/kernels/gemm/gemm.reference_output
===================================================================
--- test-suite/trunk/SingleSource/Benchmarks/Polybench/linear-algebra/kernels/gemm/gemm.reference_output
+++ test-suite/trunk/SingleSource/Benchmarks/Polybench/linear-algebra/kernels/gemm/gemm.reference_output
@@ -1 +1 @@
-622af90c831b234ea835317cd4bb4123
+d72834010eeff657031ceb2493f9e3f7
Index: test-suite/trunk/SingleSource/Benchmarks/Polybench/linear-algebra/kernels/gemm/gemm.reference_output.small
===================================================================
--- test-suite/trunk/SingleSource/Benchmarks/Polybench/linear-algebra/kernels/gemm/gemm.reference_output.small
+++ test-suite/trunk/SingleSource/Benchmarks/Polybench/linear-algebra/kernels/gemm/gemm.reference_output.small
@@ -1 +1 @@
-8685643738936fe8ee951f5e06200e56
+84e7a821aa0da2cf1d477d1b779d5e9a
Index: test-suite/trunk/SingleSource/Benchmarks/Polybench/linear-algebra/kernels/gesummv/gesummv.c
===================================================================
--- test-suite/trunk/SingleSource/Benchmarks/Polybench/linear-algebra/kernels/gesummv/gesummv.c
+++ test-suite/trunk/SingleSource/Benchmarks/Polybench/linear-algebra/kernels/gesummv/gesummv.c
@@ -50,10 +50,10 @@
 
 {
   int i;
-  char *printmat = malloc(n*8);
+  char *printmat = malloc(n*16 + 1); printmat[n*16] = 0;
 
   for (i = 0; i < n; i++)
-    print_element(y[i], i*8, printmat);
+    print_element(y[i], i*16, printmat);
   fputs(printmat, stderr);
   free(printmat);
 }
Index: test-suite/trunk/SingleSource/Benchmarks/Polybench/linear-algebra/kernels/gesummv/gesummv.reference_output
===================================================================
--- test-suite/trunk/SingleSource/Benchmarks/Polybench/linear-algebra/kernels/gesummv/gesummv.reference_output
+++ test-suite/trunk/SingleSource/Benchmarks/Polybench/linear-algebra/kernels/gesummv/gesummv.reference_output
@@ -1 +1 @@
-23b6382cb41cacbb93c669e92d7c4b7e
+a98456b34f7d94e1763bfddd510d248d
Index: test-suite/trunk/SingleSource/Benchmarks/Polybench/linear-algebra/kernels/gesummv/gesummv.reference_output.small
===================================================================
--- test-suite/trunk/SingleSource/Benchmarks/Polybench/linear-algebra/kernels/gesummv/gesummv.reference_output.small
+++ test-suite/trunk/SingleSource/Benchmarks/Polybench/linear-algebra/kernels/gesummv/gesummv.reference_output.small
@@ -1 +1 @@
-d0f71814a5109a3f373ed7587e6b0b0b
+3ab0068ba1a4869e1be8da112aef0ea5
Index: test-suite/trunk/SingleSource/Benchmarks/Polybench/linear-algebra/kernels/mvt/mvt.c
===================================================================
--- test-suite/trunk/SingleSource/Benchmarks/Polybench/linear-algebra/kernels/mvt/mvt.c
+++ test-suite/trunk/SingleSource/Benchmarks/Polybench/linear-algebra/kernels/mvt/mvt.c
@@ -50,13 +50,13 @@
 
 {
   int i;
-  char *printmat = malloc(n*8);
+  char *printmat = malloc(n*16 + 1); printmat[n*16] = 0;
 
   for (i = 0; i < n; i++)
-    print_element(x1[i], i*8, printmat);
+    print_element(x1[i], i*16, printmat);
   fputs(printmat, stderr);
   for (i = 0; i < n; i++)
-    print_element(x2[i], i*8, printmat);
+    print_element(x2[i], i*16, printmat);
   fputs(printmat, stderr);
   free(printmat);
 }
Index: test-suite/trunk/SingleSource/Benchmarks/Polybench/linear-algebra/kernels/mvt/mvt.reference_output
===================================================================
--- test-suite/trunk/SingleSource/Benchmarks/Polybench/linear-algebra/kernels/mvt/mvt.reference_output
+++ test-suite/trunk/SingleSource/Benchmarks/Polybench/linear-algebra/kernels/mvt/mvt.reference_output
@@ -1 +1 @@
-74b403cb1904ef24879eeb0f8e031b4e
+5c582f225125e1421faaf093ae45d7db
Index: test-suite/trunk/SingleSource/Benchmarks/Polybench/linear-algebra/kernels/symm/symm.c
===================================================================
--- test-suite/trunk/SingleSource/Benchmarks/Polybench/linear-algebra/kernels/symm/symm.c
+++ test-suite/trunk/SingleSource/Benchmarks/Polybench/linear-algebra/kernels/symm/symm.c
@@ -49,11 +49,11 @@
 		 DATA_TYPE POLYBENCH_2D(C,NI,NJ,ni,nj))
 {
   int i, j;
-  char *printmat = malloc(nj*8);
+  char *printmat = malloc(nj*16 + 1); printmat[nj*16] = 0;
 
   for (i = 0; i < ni; i++) {
     for (j = 0; j < nj; j++)
-      print_element(C[i][j], j*8, printmat);
+      print_element(C[i][j], j*16, printmat);
     fputs(printmat, stderr);
   }
   free(printmat);
Index: test-suite/trunk/SingleSource/Benchmarks/Polybench/linear-algebra/kernels/symm/symm.reference_output
===================================================================
--- test-suite/trunk/SingleSource/Benchmarks/Polybench/linear-algebra/kernels/symm/symm.reference_output
+++ test-suite/trunk/SingleSource/Benchmarks/Polybench/linear-algebra/kernels/symm/symm.reference_output
@@ -1 +1 @@
-e1dddd8c9c19bf9f1e7cbf0c239ce46a
+b407e999602930ed00169900fb4142de
Index: test-suite/trunk/SingleSource/Benchmarks/Polybench/linear-algebra/kernels/symm/symm.reference_output.small
===================================================================
--- test-suite/trunk/SingleSource/Benchmarks/Polybench/linear-algebra/kernels/symm/symm.reference_output.small
+++ test-suite/trunk/SingleSource/Benchmarks/Polybench/linear-algebra/kernels/symm/symm.reference_output.small
@@ -1 +1 @@
-7d05186f98518c24e49d2229c5e0ad1c
+c83a78a6235a46dd9a6fc6c7c13f3a39
Index: test-suite/trunk/SingleSource/Benchmarks/Polybench/linear-algebra/kernels/syr2k/syr2k.c
===================================================================
--- test-suite/trunk/SingleSource/Benchmarks/Polybench/linear-algebra/kernels/syr2k/syr2k.c
+++ test-suite/trunk/SingleSource/Benchmarks/Polybench/linear-algebra/kernels/syr2k/syr2k.c
@@ -49,11 +49,11 @@
 		 DATA_TYPE POLYBENCH_2D(C,NI,NI,ni,ni))
 {
   int i, j;
-  char *printmat = malloc(ni*8);
+  char *printmat = malloc(ni*16 + 1); printmat[ni*16] = 0;
 
   for (i = 0; i < ni; i++) {
     for (j = 0; j < ni; j++)
-      print_element(C[i][j], j*8, printmat);
+      print_element(C[i][j], j*16, printmat);
     fputs(printmat, stderr);
   }
   free(printmat);
Index: test-suite/trunk/SingleSource/Benchmarks/Polybench/linear-algebra/kernels/syr2k/syr2k.reference_output
===================================================================
--- test-suite/trunk/SingleSource/Benchmarks/Polybench/linear-algebra/kernels/syr2k/syr2k.reference_output
+++ test-suite/trunk/SingleSource/Benchmarks/Polybench/linear-algebra/kernels/syr2k/syr2k.reference_output
@@ -1 +1 @@
-3c957bddeddf967b7ae73a4c4546e229
+81f28343f4089dfc8545b25249067d8a
Index: test-suite/trunk/SingleSource/Benchmarks/Polybench/linear-algebra/kernels/syr2k/syr2k.reference_output.small
===================================================================
--- test-suite/trunk/SingleSource/Benchmarks/Polybench/linear-algebra/kernels/syr2k/syr2k.reference_output.small
+++ test-suite/trunk/SingleSource/Benchmarks/Polybench/linear-algebra/kernels/syr2k/syr2k.reference_output.small
@@ -1 +1 @@
-5e7018090d8a48bc5171713008ad0831
+91822c3c95174cc637351e386eb6adfa
Index: test-suite/trunk/SingleSource/Benchmarks/Polybench/linear-algebra/kernels/syrk/syrk.c
===================================================================
--- test-suite/trunk/SingleSource/Benchmarks/Polybench/linear-algebra/kernels/syrk/syrk.c
+++ test-suite/trunk/SingleSource/Benchmarks/Polybench/linear-algebra/kernels/syrk/syrk.c
@@ -46,11 +46,11 @@
 		 DATA_TYPE POLYBENCH_2D(C,NI,NI,ni,ni))
 {
   int i, j;
-  char *printmat = malloc(ni*8);
+  char *printmat = malloc(ni*16 + 1); printmat[ni*16] = 0;
 
   for (i = 0; i < ni; i++) {
     for (j = 0; j < ni; j++)
-      print_element(C[i][j], j*8, printmat);
+      print_element(C[i][j], j*16, printmat);
     fputs(printmat, stderr);
   }
   free(printmat);
Index: test-suite/trunk/SingleSource/Benchmarks/Polybench/linear-algebra/kernels/syrk/syrk.reference_output
===================================================================
--- test-suite/trunk/SingleSource/Benchmarks/Polybench/linear-algebra/kernels/syrk/syrk.reference_output
+++ test-suite/trunk/SingleSource/Benchmarks/Polybench/linear-algebra/kernels/syrk/syrk.reference_output
@@ -1 +1 @@
-622af90c831b234ea835317cd4bb4123
+d72834010eeff657031ceb2493f9e3f7
Index: test-suite/trunk/SingleSource/Benchmarks/Polybench/linear-algebra/kernels/syrk/syrk.reference_output.small
===================================================================
--- test-suite/trunk/SingleSource/Benchmarks/Polybench/linear-algebra/kernels/syrk/syrk.reference_output.small
+++ test-suite/trunk/SingleSource/Benchmarks/Polybench/linear-algebra/kernels/syrk/syrk.reference_output.small
@@ -1 +1 @@
-8685643738936fe8ee951f5e06200e56
+84e7a821aa0da2cf1d477d1b779d5e9a
Index: test-suite/trunk/SingleSource/Benchmarks/Polybench/linear-algebra/kernels/trisolv/trisolv.c
===================================================================
--- test-suite/trunk/SingleSource/Benchmarks/Polybench/linear-algebra/kernels/trisolv/trisolv.c
+++ test-suite/trunk/SingleSource/Benchmarks/Polybench/linear-algebra/kernels/trisolv/trisolv.c
@@ -55,10 +55,10 @@
 
 {
   int i;
-  char *printmat = malloc(n*8);
+  char *printmat = malloc(n*16 + 1); printmat[n*16] = 0;
 
   for (i = 0; i < n; i++)
-    print_element(x[i], i*8, printmat);
+    print_element(x[i], i*16, printmat);
   fputs(printmat, stderr);
   free(printmat);
 }
Index: test-suite/trunk/SingleSource/Benchmarks/Polybench/linear-algebra/kernels/trisolv/trisolv.reference_output
===================================================================
--- test-suite/trunk/SingleSource/Benchmarks/Polybench/linear-algebra/kernels/trisolv/trisolv.reference_output
+++ test-suite/trunk/SingleSource/Benchmarks/Polybench/linear-algebra/kernels/trisolv/trisolv.reference_output
@@ -1 +1 @@
-6b6b1c3ed0d6abeda5af3b76f89a3331
+d58d21a9e558bfff03ca8558a9d0251d
Index: test-suite/trunk/SingleSource/Benchmarks/Polybench/linear-algebra/kernels/trmm/trmm.c
===================================================================
--- test-suite/trunk/SingleSource/Benchmarks/Polybench/linear-algebra/kernels/trmm/trmm.c
+++ test-suite/trunk/SingleSource/Benchmarks/Polybench/linear-algebra/kernels/trmm/trmm.c
@@ -43,11 +43,11 @@
 		 DATA_TYPE POLYBENCH_2D(B,NI,NI,ni,ni))
 {
   int i, j;
-  char *printmat = malloc(ni*8);
+  char *printmat = malloc(ni*16 + 1); printmat[ni*16] = 0;
 
   for (i = 0; i < ni; i++) {
     for (j = 0; j < ni; j++)
-      print_element(B[i][j], j*8, printmat);
+      print_element(B[i][j], j*16, printmat);
     fputs(printmat, stderr);
   }
   free(printmat);
Index: test-suite/trunk/SingleSource/Benchmarks/Polybench/linear-algebra/kernels/trmm/trmm.reference_output
===================================================================
--- test-suite/trunk/SingleSource/Benchmarks/Polybench/linear-algebra/kernels/trmm/trmm.reference_output
+++ test-suite/trunk/SingleSource/Benchmarks/Polybench/linear-algebra/kernels/trmm/trmm.reference_output
@@ -1 +1 @@
-93f3c10749d75b23e8a19b6931b48486
+b15ac1b0d0cd75da78f553b1f9da2673
Index: test-suite/trunk/SingleSource/Benchmarks/Polybench/linear-algebra/kernels/trmm/trmm.reference_output.small
===================================================================
--- test-suite/trunk/SingleSource/Benchmarks/Polybench/linear-algebra/kernels/trmm/trmm.reference_output.small
+++ test-suite/trunk/SingleSource/Benchmarks/Polybench/linear-algebra/kernels/trmm/trmm.reference_output.small
@@ -1 +1 @@
-1d2d6c667c5f036d5e51428f5aed37ef
+7e371557248fa085c906fec361fbfeac
Index: test-suite/trunk/SingleSource/Benchmarks/Polybench/linear-algebra/solvers/lu/lu.c
===================================================================
--- test-suite/trunk/SingleSource/Benchmarks/Polybench/linear-algebra/solvers/lu/lu.c
+++ test-suite/trunk/SingleSource/Benchmarks/Polybench/linear-algebra/solvers/lu/lu.c
@@ -39,11 +39,11 @@
 
 {
   int i, j;
-  char *printmat = malloc(n*8);
+  char *printmat = malloc(n*16 + 1); printmat[n*16] = 0;
 
   for (i = 0; i < n; i++) {
     for (j = 0; j < n; j++)
-      print_element(A[i][j], j*8, printmat);
+      print_element(A[i][j], j*16, printmat);
     fputs(printmat, stderr);
   }
   free(printmat);
Index: test-suite/trunk/SingleSource/Benchmarks/Polybench/linear-algebra/solvers/lu/lu.reference_output
===================================================================
--- test-suite/trunk/SingleSource/Benchmarks/Polybench/linear-algebra/solvers/lu/lu.reference_output
+++ test-suite/trunk/SingleSource/Benchmarks/Polybench/linear-algebra/solvers/lu/lu.reference_output
@@ -1 +1 @@
-83a3b7d418f8e147ab88fa04966c7c37
+064a3e60f06aeecd598f7f66b60035b6
Index: test-suite/trunk/SingleSource/Benchmarks/Polybench/medley/floyd-warshall/floyd-warshall.c
===================================================================
--- test-suite/trunk/SingleSource/Benchmarks/Polybench/medley/floyd-warshall/floyd-warshall.c
+++ test-suite/trunk/SingleSource/Benchmarks/Polybench/medley/floyd-warshall/floyd-warshall.c
@@ -39,11 +39,11 @@
 
 {
   int i, j;
-  char *printmat = malloc(n*8);
+  char *printmat = malloc(n*16 + 1); printmat[n*16] = 0;
 
   for (i = 0; i < n; i++) {
     for (j = 0; j < n; j++)
-      print_element(path[i][j], j*8, printmat);
+      print_element(path[i][j], j*16, printmat);
     fputs(printmat, stderr);
   }
   free(printmat);
Index: test-suite/trunk/SingleSource/Benchmarks/Polybench/medley/floyd-warshall/floyd-warshall.reference_output
===================================================================
--- test-suite/trunk/SingleSource/Benchmarks/Polybench/medley/floyd-warshall/floyd-warshall.reference_output
+++ test-suite/trunk/SingleSource/Benchmarks/Polybench/medley/floyd-warshall/floyd-warshall.reference_output
@@ -1 +1 @@
-72d6ab856485d06c244c238122cd13a8
+cb96c8306dcb184cd35dea4694abb3ab
Index: test-suite/trunk/SingleSource/Benchmarks/Polybench/medley/floyd-warshall/floyd-warshall.reference_output.small
===================================================================
--- test-suite/trunk/SingleSource/Benchmarks/Polybench/medley/floyd-warshall/floyd-warshall.reference_output.small
+++ test-suite/trunk/SingleSource/Benchmarks/Polybench/medley/floyd-warshall/floyd-warshall.reference_output.small
@@ -1 +1 @@
-90b8c539c785cf4ca34231d074058d43
+88615b373db5ad02fe09a10d3bba92dc
Index: test-suite/trunk/SingleSource/Benchmarks/Polybench/stencils/adi/adi.c
===================================================================
--- test-suite/trunk/SingleSource/Benchmarks/Polybench/stencils/adi/adi.c
+++ test-suite/trunk/SingleSource/Benchmarks/Polybench/stencils/adi/adi.c
@@ -45,11 +45,11 @@
 
 {
   int i, j;
-  char *printmat = malloc(n*8);
+  char *printmat = malloc(n*16 + 1); printmat[n*16] = 0;
 
   for (i = 0; i < n; i++) {
     for (j = 0; j < n; j++)
-      print_element(X[i][j], j*8, printmat);
+      print_element(X[i][j], j*16, printmat);
     fputs(printmat, stderr);
   }
   free(printmat);
Index: test-suite/trunk/SingleSource/Benchmarks/Polybench/stencils/adi/adi.reference_output
===================================================================
--- test-suite/trunk/SingleSource/Benchmarks/Polybench/stencils/adi/adi.reference_output
+++ test-suite/trunk/SingleSource/Benchmarks/Polybench/stencils/adi/adi.reference_output
@@ -1 +1 @@
-1823d3f6cf83a9b28749d6edaa5fedc9
+a954c885e6119039294e81fd5cda4324
Index: test-suite/trunk/SingleSource/Benchmarks/Polybench/stencils/adi/adi.reference_output.small
===================================================================
--- test-suite/trunk/SingleSource/Benchmarks/Polybench/stencils/adi/adi.reference_output.small
+++ test-suite/trunk/SingleSource/Benchmarks/Polybench/stencils/adi/adi.reference_output.small
@@ -1 +1 @@
-5182fd4f225bdeea7e9cb31766de5ee2
+dac9b9113be723b7da52c2887d065b04
Index: test-suite/trunk/SingleSource/Benchmarks/Polybench/stencils/fdtd-2d/fdtd-2d.c
===================================================================
--- test-suite/trunk/SingleSource/Benchmarks/Polybench/stencils/fdtd-2d/fdtd-2d.c
+++ test-suite/trunk/SingleSource/Benchmarks/Polybench/stencils/fdtd-2d/fdtd-2d.c
@@ -52,17 +52,17 @@
 		 DATA_TYPE POLYBENCH_2D(hz,NX,NY,nx,ny))
 {
   int i, j;
-  char *printmat = malloc(ny*8);
+  char *printmat = malloc(ny*16 + 1); printmat[ny*16] = 0;
 
   for (i = 0; i < nx; i++) {
     for (j = 0; j < ny; j++)
-      print_element(ex[i][j], j*8, printmat);
+      print_element(ex[i][j], j*16, printmat);
     fputs(printmat, stderr);
     for (j = 0; j < ny; j++)
-      print_element(ey[i][j], j*8, printmat);
+      print_element(ey[i][j], j*16, printmat);
     fputs(printmat, stderr);
     for (j = 0; j < ny; j++)
-      print_element(hz[i][j], j*8, printmat);
+      print_element(hz[i][j], j*16, printmat);
     fputs(printmat, stderr);
   }
   free(printmat);
Index: test-suite/trunk/SingleSource/Benchmarks/Polybench/stencils/fdtd-2d/fdtd-2d.reference_output
===================================================================
--- test-suite/trunk/SingleSource/Benchmarks/Polybench/stencils/fdtd-2d/fdtd-2d.reference_output
+++ test-suite/trunk/SingleSource/Benchmarks/Polybench/stencils/fdtd-2d/fdtd-2d.reference_output
@@ -1 +1 @@
-70ea9d7528f3022d056656af4d540929
+2eef9f06f24753aeca4fdb6190dfca15
Index: test-suite/trunk/SingleSource/Benchmarks/Polybench/stencils/fdtd-2d/fdtd-2d.reference_output.small
===================================================================
--- test-suite/trunk/SingleSource/Benchmarks/Polybench/stencils/fdtd-2d/fdtd-2d.reference_output.small
+++ test-suite/trunk/SingleSource/Benchmarks/Polybench/stencils/fdtd-2d/fdtd-2d.reference_output.small
@@ -1 +1 @@
-cd41e645df13df9fa373727b6e285f64
+cf61a62ee26fc0ece455545f5789a4d8
Index: test-suite/trunk/SingleSource/Benchmarks/Polybench/stencils/jacobi-1d-imper/jacobi-1d-imper.c
===================================================================
--- test-suite/trunk/SingleSource/Benchmarks/Polybench/stencils/jacobi-1d-imper/jacobi-1d-imper.c
+++ test-suite/trunk/SingleSource/Benchmarks/Polybench/stencils/jacobi-1d-imper/jacobi-1d-imper.c
@@ -42,10 +42,10 @@
 
 {
   int i;
-  char *printmat = malloc(n*8);
+  char *printmat = malloc(n*16 + 1); printmat[n*16] = 0;
 
   for (i = 0; i < n; i++)
-    print_element(A[i], i*8, printmat);
+    print_element(A[i], i*16, printmat);
   fputs(printmat, stderr);
   free(printmat);
 }
Index: test-suite/trunk/SingleSource/Benchmarks/Polybench/stencils/jacobi-1d-imper/jacobi-1d-imper.reference_output
===================================================================
--- test-suite/trunk/SingleSource/Benchmarks/Polybench/stencils/jacobi-1d-imper/jacobi-1d-imper.reference_output
+++ test-suite/trunk/SingleSource/Benchmarks/Polybench/stencils/jacobi-1d-imper/jacobi-1d-imper.reference_output
@@ -1 +1 @@
-53110044128f4550e864b8dd19771ba2
+a9f60f2ecfde681ef37298d993dfb0db
Index: test-suite/trunk/SingleSource/Benchmarks/Polybench/stencils/jacobi-2d-imper/jacobi-2d-imper.c
===================================================================
--- test-suite/trunk/SingleSource/Benchmarks/Polybench/stencils/jacobi-2d-imper/jacobi-2d-imper.c
+++ test-suite/trunk/SingleSource/Benchmarks/Polybench/stencils/jacobi-2d-imper/jacobi-2d-imper.c
@@ -43,11 +43,11 @@
 
 {
   int i, j;
-  char *printmat = malloc(n*8);
+  char *printmat = malloc(n*16 + 1); printmat[n*16] = 0;
 
   for (i = 0; i < n; i++) {
     for (j = 0; j < n; j++)
-      print_element(A[i][j], j*8, printmat);
+      print_element(A[i][j], j*16, printmat);
     fputs(printmat, stderr);
   }
   free(printmat);
Index: test-suite/trunk/SingleSource/Benchmarks/Polybench/stencils/jacobi-2d-imper/jacobi-2d-imper.reference_output
===================================================================
--- test-suite/trunk/SingleSource/Benchmarks/Polybench/stencils/jacobi-2d-imper/jacobi-2d-imper.reference_output
+++ test-suite/trunk/SingleSource/Benchmarks/Polybench/stencils/jacobi-2d-imper/jacobi-2d-imper.reference_output
@@ -1 +1 @@
-58213157f3b250a08e33f260ecf70ee0
+2f14483f199678d07eb29cc539b07546
Index: test-suite/trunk/SingleSource/Benchmarks/Polybench/stencils/jacobi-2d-imper/jacobi-2d-imper.reference_output.small
===================================================================
--- test-suite/trunk/SingleSource/Benchmarks/Polybench/stencils/jacobi-2d-imper/jacobi-2d-imper.reference_output.small
+++ test-suite/trunk/SingleSource/Benchmarks/Polybench/stencils/jacobi-2d-imper/jacobi-2d-imper.reference_output.small
@@ -1 +1 @@
-6a17fdd70db1d0cab444fdb6e69c1363
+448bf8d2e52eb27f7f8d13ce2d7760af
Index: test-suite/trunk/SingleSource/Benchmarks/Polybench/stencils/seidel-2d/seidel-2d.c
===================================================================
--- test-suite/trunk/SingleSource/Benchmarks/Polybench/stencils/seidel-2d/seidel-2d.c
+++ test-suite/trunk/SingleSource/Benchmarks/Polybench/stencils/seidel-2d/seidel-2d.c
@@ -39,11 +39,11 @@
 
 {
   int i, j;
-  char *printmat = malloc(n*8);
+  char *printmat = malloc(n*16 + 1); printmat[n*16] = 0;
 
   for (i = 0; i < n; i++) {
     for (j = 0; j < n; j++)
-      print_element(A[i][j], j*8, printmat);
+      print_element(A[i][j], j*16, printmat);
     fputs(printmat, stderr);
   }
   free(printmat);
Index: test-suite/trunk/SingleSource/Benchmarks/Polybench/stencils/seidel-2d/seidel-2d.reference_output
===================================================================
--- test-suite/trunk/SingleSource/Benchmarks/Polybench/stencils/seidel-2d/seidel-2d.reference_output
+++ test-suite/trunk/SingleSource/Benchmarks/Polybench/stencils/seidel-2d/seidel-2d.reference_output
@@ -1 +1 @@
-58213157f3b250a08e33f260ecf70ee0
+0769894932e6a061493d271cc177d667
Index: test-suite/trunk/SingleSource/Benchmarks/Polybench/stencils/seidel-2d/seidel-2d.reference_output.small
===================================================================
--- test-suite/trunk/SingleSource/Benchmarks/Polybench/stencils/seidel-2d/seidel-2d.reference_output.small
+++ test-suite/trunk/SingleSource/Benchmarks/Polybench/stencils/seidel-2d/seidel-2d.reference_output.small
@@ -1 +1 @@
-6a17fdd70db1d0cab444fdb6e69c1363
+de38fe7aea37fb6a05cb2f18bff8502d
Index: test-suite/trunk/SingleSource/Benchmarks/Polybench/utilities/polybench.h
===================================================================
--- test-suite/trunk/SingleSource/Benchmarks/Polybench/utilities/polybench.h
+++ test-suite/trunk/SingleSource/Benchmarks/Polybench/utilities/polybench.h
@@ -610,24 +610,32 @@
    for a long time), we split the output into an encoded string,
    and print it as a simple char pointer, M times. */
 static inline
-void print_element(float el, int pos, char *out)
+void print_element(double el, int pos, char *out)
 {
   union {
-    float datum;
-    char bytes[4];
+    double datum;
+    char bytes[8];
   } block;
 
   block.datum = el;
   /* each nibble as a char, within the printable range */
 #ifdef __BIG_ENDIAN__
-  *(out+pos+7) = (block.bytes[0]&0xF0>>4)+'0';
-  *(out+pos+6) = (block.bytes[0]&0x0F)   +'0';
-  *(out+pos+5) = (block.bytes[1]&0xF0>>4)+'0';
-  *(out+pos+4) = (block.bytes[1]&0x0F)   +'0';
-  *(out+pos+3) = (block.bytes[2]&0xF0>>4)+'0';
-  *(out+pos+2) = (block.bytes[2]&0x0F)   +'0';
-  *(out+pos+1) = (block.bytes[3]&0xF0>>4)+'0';
-  *(out+pos) =   (block.bytes[3]&0x0F)   +'0';
+  *(out+pos+15) = (block.bytes[0]&0xF0>>4)+'0';
+  *(out+pos+14) = (block.bytes[0]&0x0F)   +'0';
+  *(out+pos+13) = (block.bytes[1]&0xF0>>4)+'0';
+  *(out+pos+12) = (block.bytes[1]&0x0F)   +'0';
+  *(out+pos+11) = (block.bytes[2]&0xF0>>4)+'0';
+  *(out+pos+10) = (block.bytes[2]&0x0F)   +'0';
+  *(out+pos+9) = (block.bytes[3]&0xF0>>4)+'0';
+  *(out+pos+8) =   (block.bytes[3]&0x0F)   +'0';
+  *(out+pos+7) = (block.bytes[4]&0xF0>>4)+'0';
+  *(out+pos+6) = (block.bytes[4]&0x0F)   +'0';
+  *(out+pos+5) = (block.bytes[5]&0xF0>>4)+'0';
+  *(out+pos+4) = (block.bytes[5]&0x0F)   +'0';
+  *(out+pos+3) = (block.bytes[6]&0xF0>>4)+'0';
+  *(out+pos+2) = (block.bytes[6]&0x0F)   +'0';
+  *(out+pos+1) = (block.bytes[7]&0xF0>>4)+'0';
+  *(out+pos) =   (block.bytes[7]&0x0F)   +'0';
 #else
   *(out+pos)   = (block.bytes[0]&0xF0>>4)+'0';
   *(out+pos+1) = (block.bytes[0]&0x0F)   +'0';
@@ -637,6 +645,14 @@
   *(out+pos+5) = (block.bytes[2]&0x0F)   +'0';
   *(out+pos+6) = (block.bytes[3]&0xF0>>4)+'0';
   *(out+pos+7) = (block.bytes[3]&0x0F)   +'0';
+  *(out+pos+8) = (block.bytes[4]&0xF0>>4)+'0';
+  *(out+pos+9) = (block.bytes[4]&0x0F)   +'0';
+  *(out+pos+10) = (block.bytes[5]&0xF0>>4)+'0';
+  *(out+pos+11) = (block.bytes[5]&0x0F)   +'0';
+  *(out+pos+12) = (block.bytes[6]&0xF0>>4)+'0';
+  *(out+pos+13) = (block.bytes[6]&0x0F)   +'0';
+  *(out+pos+14) = (block.bytes[7]&0xF0>>4)+'0';
+  *(out+pos+15) = (block.bytes[7]&0x0F)   +'0';
 #endif
 }