Clean up canonical mt benchmarks and reorganize extra versions in /mt. All versions support support at least 1/2/4 threads.

1 files changed, 37 insertions, 0 deletions

diff --git a/mt/ad_matmul.c b/mt/ad_matmul.c
new file mode 100755
index 0000000..60e6e6c
--- /dev/null
+++ b/mt/ad_matmul.c
@@ -0,0 +1,37 @@
+#include "stdlib.h"
+
+#include "util.h"
+
+#include "dataset.h"
+void __attribute__((noinline)) matmul(const int coreid, const int ncores, const int lda,  const data_t A[], const data_t B[], data_t C[] )
+{
+   int i, k;
+   int j = coreid*(lda/ncores);
+   int jend = (coreid+1)*(lda/ncores);
+   for ( ; j < jend; j++ )
+   {
+      int j32 = j << 5;
+      data_t* Cj32 = C + j32;
+      for ( k = 0; k < 32; k+=2 )
+      {
+	 data_t Aj32k  = A[k + j32];
+	 data_t Aj32k2 = A[k + 1 + j32];
+	 data_t* Bk32  = B + (k << 5);
+	 data_t* Bk322 = Bk32 + 32;
+	 for ( i = 0; i < 32; i+=4 )
+	 {
+            Cj32[i]   += Aj32k  * Bk32   [i];
+            Cj32[i]   += Aj32k2 * Bk322  [i];
+            Cj32[i+1] += Aj32k  * Bk32 [i+1];
+            Cj32[i+1] += Aj32k2 * Bk322[i+1];
+            Cj32[i+2] += Aj32k  * Bk32 [i+2];
+            Cj32[i+2] += Aj32k2 * Bk322[i+2];
+            Cj32[i+3] += Aj32k  * Bk32 [i+3];
+            Cj32[i+3] += Aj32k2 * Bk322[i+3];
+	 }
+         barrier(ncores);
+      }
+   }
+   
+ 
+}