[Assorted-commits] SF.net SVN: assorted:[1230] sandbox/trunk/src/cc/copy.cc

[<yan...@us...>]

Revision: 1230
          http://assorted.svn.sourceforge.net/assorted/?rev=1230&view=rev
Author:   yangzhang
Date:     2009-02-24 20:03:15 +0000 (Tue, 24 Feb 2009)

Log Message:
-----------
added trial at beating memcpy

Added Paths:
-----------
    sandbox/trunk/src/cc/copy.cc

Added: sandbox/trunk/src/cc/copy.cc
===================================================================
--- sandbox/trunk/src/cc/copy.cc	                        (rev 0)
+++ sandbox/trunk/src/cc/copy.cc	2009-02-24 20:03:15 UTC (rev 1230)
@@ -0,0 +1,100 @@
+// Just can't beat memcpy!
+#include <cstring>
+#include <algorithm>
+#include <commons/time.h>
+using namespace commons;
+using namespace std;
+enum { size = (int)200e6, reps = 5 };
+void mydumbmemcpy(char *q, char *p, size_t size) {
+  for (size_t i = 0; i < size; ++i)
+    q[i] = p[i];
+}
+void mycopy(char *p, char *end, char *q) {
+  for (; p < end; ++p, ++q)
+    *q = *p;
+}
+void mymemcpy(char *q, char *p, size_t size) {
+  long long *qq = reinterpret_cast<long long*>(q),
+      *pp = reinterpret_cast<long long*>(p);
+  size /= sizeof(long long);
+  for (size_t i = 0; i < size; ++i)
+    qq[i] = pp[i];
+}
+void unrolled(char *q, char *p, size_t size) {
+  long long *qq = reinterpret_cast<long long*>(q),
+            *pp = reinterpret_cast<long long*>(p),
+            *end = qq + size / sizeof(long long);
+  while (qq < end) {
+    *qq++ = *pp++;
+    *qq++ = *pp++;
+    *qq++ = *pp++;
+    *qq++ = *pp++;
+    *qq++ = *pp++;
+    *qq++ = *pp++;
+    *qq++ = *pp++;
+    *qq++ = *pp++;
+  }
+}
+//void mycopy(char *p, char *end, char *q) {
+//  for (; p < end; ++p, ++q)
+//    *q = *p;
+//}
+int main() {
+  char *p = new char[size], *q = new char[size];
+  cout << "memcpy" << endl;
+  for (int i = 0; i < reps; ++i) {
+    long long start = current_time_millis();
+    memcpy(q, p, size);
+    long long diff = current_time_millis() - start;
+    cout << diff << " ms or " << size / double(diff) / 1000 << " MB/s" << endl;
+  }
+  cout << "copy" << endl;
+  for (int i = 0; i < reps; ++i) {
+    long long start = current_time_millis();
+    copy(p, p + size, q);
+    long long diff = current_time_millis() - start;
+    cout << diff << " ms or " << size / double(diff) / 1000 << " MB/s" << endl;
+  }
+  cout << "unrolled" << endl;
+  for (int i = 0; i < reps; ++i) {
+    long long start = current_time_millis();
+    unrolled(q, p, size);
+    long long diff = current_time_millis() - start;
+    cout << diff << " ms or " << size / double(diff) / 1000 << " MB/s" << endl;
+  }
+  cout << "mydumbmemcpy" << endl;
+  for (int i = 0; i < reps; ++i) {
+    long long start = current_time_millis();
+    mydumbmemcpy(q, p, size);
+    long long diff = current_time_millis() - start;
+    cout << diff << " ms or " << size / double(diff) / 1000 << " MB/s" << endl;
+  }
+  cout << "mycopy" << endl;
+  for (int i = 0; i < reps; ++i) {
+    long long start = current_time_millis();
+    mycopy(p, p + size, q);
+    long long diff = current_time_millis() - start;
+    cout << diff << " ms or " << size / double(diff) / 1000 << " MB/s" << endl;
+  }
+  cout << "mymemcpy" << endl;
+  for (int i = 0; i < reps; ++i) {
+    long long start = current_time_millis();
+    mymemcpy(q, p, size);
+    long long diff = current_time_millis() - start;
+    cout << diff << " ms or " << size / double(diff) / 1000 << " MB/s" << endl;
+  }
+  return 0;
+}
+#if 0
+482 ms or 414.938 MB/s
+122 ms or 1639.34 MB/s
+117 ms or 1709.4 MB/s
+120 ms or 1666.67 MB/s
+118 ms or 1694.92 MB/s
+--
+165 ms or 1212.12 MB/s
+162 ms or 1234.57 MB/s
+152 ms or 1315.79 MB/s
+153 ms or 1307.19 MB/s
+153 ms or 1307.19 MB/s
+#endif


This was sent by the SourceForge.net collaborative development platform, the world's largest Open Source development site.