1 files changed, 57 insertions, 0 deletions
diff --git a/xddm/display/res.c b/xddm/display/res.c
index 60e9bcb..589218b 100644
--- a/xddm/display/res.c
+++ b/xddm/display/res.c
@@ -1283,6 +1283,63 @@ static _inline void fast_memcpy_unaligment(void *dest, const void *src, size_t l
 
 #endif
 
+uint64_t time_usecs(void)
+{
+  ENG_TIME_FIELDS systime;
+  EngQueryLocalTime(&systime);
+  return (uint64_t)(systime.usMilliseconds * 1000 + systime.usSecond * 1e6 +
+                     systime.usMinute * 60e6 + systime.usHour * 3600e6);
+}
+
+void benchmark_memcpy(PDev *pdev)
+{
+  size_t i;
+  unsigned char *src_unaligned;
+  unsigned char *dest_unaligned;
+  uint64_t start, total1, total2;
+  unsigned char *src = NULL;
+  unsigned char *dest = NULL;
+  size_t size = 1024;
+  size_t iter = 1024 * 1024;
+
+  for (size = 1024; size < 1024*1024*2; size *= 2, iter /= 2) {
+    src_unaligned = EngAllocMem(0, size + 31, ALLOC_TAG);
+    dest_unaligned = EngAllocMem(0, size + 31, ALLOC_TAG);
+    src = (unsigned char *)((size_t)(src_unaligned + 31) & ~0x1f);
+    dest = (unsigned char *)((size_t)(dest_unaligned + 31) & ~0x1f);
+
+    for (i = 0 ; i < size ; ++i)
+      src[i] = i;
+
+    start = time_usecs();
+    for (i = 0 ; i < iter ; ++i) {
+      fast_memcpy_aligned(dest, src, size);
+    }
+    total2 = time_usecs() - start;
+
+    {
+      int errors = 0;
+      for (i = 0 ; i < size ; ++i) {
+        if (dest[i] != src[i]) {
+          errors++;
+        }
+      }
+      if (errors > 0) {
+        DEBUG_PRINT((pdev, 1, "!!! copy errors %d !!!\n", errors));
+      }
+    }
+
+    start = time_usecs();
+    for (i = 0 ; i < iter ; ++i)
+      memcpy(dest, src, size);
+    total1 = time_usecs() - start;
+
+    DEBUG_PRINT((pdev, 1, "%d: %lld, %lld\n", size, total1, total2));
+    EngFreeMem(src_unaligned);
+    EngFreeMem(dest_unaligned);
+  }
+}
+
 #ifdef DBG
     #define PutBytesAlign __PutBytesAlign
 #define PutBytes(pdev, chunk, now, end, src, size, page_counter, alloc_size, use_sse)\