Mark RV32 tests as such

[riscv-tests.git] / benchmarks / mm / mm_main.c

diff --git a/benchmarks/mm/mm_main.c b/benchmarks/mm/mm_main.c
index 522768a4fb11276e4145dc1cf1ed58867c816afb..133b5a2d73a17434e8aefef7d8e4ebf4ba8d402e 100644 (file)
--- a/benchmarks/mm/mm_main.c
+++ b/benchmarks/mm/mm_main.c
@@ -1,22 +1,22 @@
+// See LICENSE for license details.
+
 #include "common.h"
 #include <assert.h>
 #include <stdlib.h>
+#include <stdio.h>
 #include "util.h"
 
+#pragma GCC optimize ("unroll-loops")
+
 void thread_entry(int cid, int nc)
 {
   const int R = 8;
   int m, n, p;
+  uint64_t s = 0xdeadbeefU;
   
-  if (have_vec) {
-    m = HCBM;
-    n = HCBN;
-    p = HCBK;
-  } else {
-    m = CBM;
-    n = CBN;
-    p = CBK;
-  }
+  m = CBM;
+  n = CBN;
+  p = CBK;
 
   t a[m*p];
   t b[p*n];
@@ -24,33 +24,24 @@ void thread_entry(int cid, int nc)
 
   for (size_t i = 0; i < m; i++)
     for (size_t j = 0; j < p; j++)
-      a[i*p+j] = i+j;
+      a[i*p+j] = (t)(s = lfsr(s));
   for (size_t i = 0; i < p; i++)
     for (size_t j = 0; j < n; j++)
-      b[i*n+j] = i-j;
+      b[i*n+j] = (t)(s = lfsr(s));
   memset(c, 0, m*n*sizeof(c[0]));
 
   size_t instret, cycles;
-  if (have_vec) {
-    for (int i = 0; i < R; i++)
-    {
-      instret = -rdinstret();
-      cycles = -rdcycle();
-      mm_rb_hwacha(m, n, p, a, p, b, n, c, n);
-      instret += rdinstret();
-      cycles += rdcycle();
-    }
-  } else {
-    for (int i = 0; i < R; i++)
-    {
-      instret = -rdinstret();
-      cycles = -rdcycle();
-      mm(m, n, p, a, p, b, n, c, n);
-      instret += rdinstret();
-      cycles += rdcycle();
-    }
+  for (int i = 0; i < R; i++)
+  {
+    instret = -read_csr(minstret);
+    cycles = -read_csr(mcycle);
+    mm(m, n, p, a, p, b, n, c, n);
+    instret += read_csr(minstret);
+    cycles += read_csr(mcycle);
   }
 
+  asm volatile("fence");
+
   printf("C%d: reg block %dx%dx%d, cache block %dx%dx%d\n",
          cid, RBM, RBN, RBK, CBM, CBN, CBK);
   printf("C%d: %d instructions\n", cid, (int)(instret));
@@ -64,9 +55,10 @@ void thread_entry(int cid, int nc)
     for (size_t j = 0; j < n; j++)
     {
       t s = 0;
-      for (size_t aik = i, bkj = -j; aik < i+p; aik++, bkj++)
-        s += (t)aik*(t)bkj;
-      if (fabs(c[i*n+j]-s*R) > 1e-6*s)
+      for (size_t k = 0; k < p; k++)
+        s += a[i*p+k] * b[k*n+j];
+      s *= R;
+      if (fabs(c[i*n+j]-s) > fabs(1e-6*s))
       {
         printf("C%d: c[%lu][%lu] %f != %f\n", cid, i, j, c[i*n+j], s);
         exit(1);